Mortal Kombat Hipsters

azizisbored:

This Between Two Ferns with Steve Carrell is hysterical.

Just watched this at work on my phone, was laughing out loud, drew many disapproving looks.

I finally got my act together and got the internet, I only signed up and bought the router 2 months ago. So thanks Bigdaddy and default, you were slow and unreliable, but it was better then nothing. If you want me to show you how to password protect your wi fi, just let me know.

SELL!! SELL!!! SELL!!!

Mombasa - Hans Zimmer, Johnny Marr, etc Performed Live at Inception Premiere

Interesting how the big composers of our day are the ones scoring futuristic sci-fi movies. Imagine how epic a movie would have been if Beethoven or Mozart had scored it.

Points for Johnny Marr, the bearded drummer and the asian chick rocking the shit out of the violin.

I’ll have what he’s having. Ami’right?!

Someone was so affected by Murphy’s death that they wrote this on the bottom (!!!) of a toilet seat at a dirty rock club. That’s star power.

Yo, if you read any reports about shingles and fiberglass and debris and shit lying all over Sherbourne St., it’s probably because Sleigh Bells and Die Antwoord blew the fucking roof off the fucking Phoenix Concert Theatre on Tuesday night….

Via. Now Toronto

Bill Murray: [discussing NBC’s current Thursday night comedy block] I’m out of touch. I have no idea. I never saw the original Office. I never saw this Office. I never even saw Clerks. Like I never saw, what’s-his-name, Larry David’s show.

GQ: Curb Your Enthusiasm?

Bill Murray: No! The…

LCD Soundsystem- All My Friends

• 8x what the CEO makes in a year
• Just less then one of their charitable donations
• 1/10th of the gain that GS’s stock had in after hours trading
• 2 weeks worth of profit

Full story

I have watched a heck of a lot of Top Chef in my day, even some Top Model… but this, TOP SHOT, is clearly the best of the genre.

Cool pics from the Tour de France. Last summer I watched a lot of the Tour and really enjoyed it. Stupid career.

Spain hasn’t stolen this much gold since they met the native Americans.

Fo’ Drizzle

Tomorrow is that day from Back To The Future!

From The Star this morning…

Last word goes to an English fan overheard in Durban:

“The World Cup has turned out like World War II. The French surrendered early, the USA arrived late, and we’re left to fight the Germans.”

Cop car is on fire on Queen st. this far from my balcony.

Protest going up spadina from queen

Update: the police pushed protesters up to within a block of my apartment. There are more photographers and jerks like me gawking then protesters. S pretty ridiculous event by all accounts

Football-Fever hits G8 meeting

Nice to read some G8/20 news that isn’t doom and gloom or protest related.

It is about how Germany and England are playing on Sunday in the quarter finals and how the leaders are planning to sneak away to watch the game.

Damn you G20!!!!!

Thank you internet for making my day.

via. collegehumor

Has an evil mastermind created some sort of seismological super weapon? If there is a thunderstorm tonight, I am getting as far away from this city as possible…

The G20 fence has gone up outside my office. We are on the unprotected side… Should get pretty protesty

Sleigh Bells - Kids
A fun game I like to play at work is to drink a liter of coffee, while doing monotonous excel work. I then put this album on and try not to jump on my desk and dance-fight my co-workers.
Be sure to turn up the base, because this cd is a banger!

As I walked to work this morning I peered down an alley only to see a middle-aged black man riding his bike backwards (the bike going forward as per usual, him sitting on the bike backwards) without a care in the world. Then I sat in front of a computer for 9 hours. Who’s the crazy person?

Friday’s job report was awful. For most new high school and college grads finding a job is harder than ever. Meanwhile, states are cutting summer jobs for disadvantaged young people. What to do with this army of young unemployed? Send them to the Gulf to clean up beaches and wetlands, and send the bill to BP.

Florida’s panhandle beaches are already marred with sticky brown globs of oil. Workers with blue rubber gloves and plastic bags are already losing the battle to keep them clean. Pelicans and other wildlife coated in oil tar are dying by the droves.

It will get far worse. Most of the oil hasn’t hit land yet. When it does, hundreds of thousands of workers will be needed to clean beaches, siphon off oil from wetlands, and rescue stranded wildlife. Tens of thousands more will have to bring in new landfill, replace tarred sea walls, and rebuild shoreline infrastructure.

Yet we’ve got hundreds of thousands of young people sitting on their hands right now because they can’t find jobs. Many are from affected coastal areas, where the tourist and fishing industries have been decimated by the spill.

The President should order BP to establish a $5 billion clean-up fund, and immediately put America’s army of unemployed young people to work saving the Gulf coast. Call it the new Civilian Conservation Corps.

(The old CCC — created by FDR at another time of massive unemployment and environmental stress — gave millions of young Americans jobs and training to reforest lands that had been degraded, provide emergency flood relief in the Ohio and Mississippi valleys, and build the infrastructure for our national parks.)

I was at OYS’s sail past this weekend, as Paul was the Commodore. He started the day with a monster shot of Scotch. Then we sailed out with the piper blaring and everyone saluting. The weather turned out great, the party was a lot of fun (I left when my parents started dancing- no one needs to see that) and everyone seemed to have a good time. Another nice weekend.

Cancellation comes on the heels of cancellations by Klaxons and Gorillaz Sound System in wake of Gaza flotilla raid.

Oh sure, blame the tea

This is awesome

I had a pretty glorious Sunday. It included a brun (brie run), beers in the park, a slightly ridiculous tight rope walker (above), great friends and gourmet burgers.

Unfortunately, such a good weekend always leads to a gloomy Monday.

So I was quietly watching law and order last night when i heard what I thought was gun shots (though having never heard gunshots i was unsure). I went out to the balcony to see what was up, where I then saw 3 young dudes run by, with gloves on and something in their hands. They got into a car and peeled out.

The ensuing police investigation confirmed what I saw. There were no ambulances so I don’t think anyone got hurt. Toronto is a lot different from Copenhagen…

Sitting on my balcony having an espresso and listening to the new LCD Soundsystem… Long weekends are great!

This Movie Is Broken - Trailer

Seems pretty hipster-ish, but the Broken Social Scene stuff looks pretty cool

Woken up to a fire in the apartment two over from mine. Good thing the whole building is made of cement.

The best part is a got an insurance quote on Monday but haven’t followed up

YUP

yup

Stress Testing with a Macroeconomic Credit Risk Model for the Corporate Loan Portfolio of Canadian Banks

This paper focuses on the effects of macroeconomic shocks on the credit risk of corporate loans in the Canadian banking industry. To do this I have created an aggregate corporate loan portfolio by combining data on default rates of each industry in Canada. The default rates of the portfolio are used as a function of the macroeconomic environment, which are then subjected to stress scenarios. Stress testing is the process of determining performance of an entity during extreme yet not implausible events. The purpose of which is to expose any weaknesses that may exist.

This paper follows the methods of Virolainen (2004) and Misina, Tessier and Dey (2006), which incorporate similar tests in Finland and Canada respectively. While Misina et al. also use Canadian data, they perform their test on a national scale where I have performed my test on regional scale (i.e. for each province in Canada). The reason I have chosen to do so, is because the extreme size and diversity of Canada. I wanted to see if the stress tests could uncover particular regional weaknesses, while at the same time creating a more accurate corporate loan portfolio for Canada in aggregate.

The credit risk model I use explicitly incorporates macroeconomic variables (GDP growth, real interest rates for Canada and the US and the commodity price index) with the appropriate lags (determined by using the Akaike information criterion). Also incorporated is a static loss given default rate, sectoral exposure rate (amount of corporate loans going to a specific industry) and regional exposure rate (amount of corporate loans going to a specific province). I have done so to most accurately define the Canadian corporate loan portfolio as possible.

Once the model was properly specified, I used a Monte Carlo simulation to forecast the expected loan losses for 8 periods (2 years) and repeated this process 5000 times to give me the corresponding loan loss distribution. I then created shocks involving dramatic GDP growth rate decreases, real interest rate increases, commodity price index increases and finally a combination dooms‐day scenario (which incorporated a GDP growth rate decrease and real interest rate increases over several periods) and ran the simulation again.

The results that followed were mostly as predicted. I provided both expected loss and the 99% value at risk estimates. I provided results after the 4th quarter (1 year) and 8th quarter (2 years) for each province and Canada on aggregate, as well as a graphical representation of the loan loss distribution for Canada.

Overall, the Canadian economy faired well, though there were definite differences amongst the provinces. The GDP shock was the most severe, especially on the west coast. Real interest rate increases in Canada were not as dramatic as expected given the severity of the shock. This time the east coast proved far more vulnerable to increased losses. An increase to US real interest rates was the most finicky, creating small increases to losses in the east, but decreases in the west. The commodity price index shock was the least severe across the nation, especially considering the size of the shock introduced. The combination scenario confirmed Alberta and Quebec as the most fragile provinces during our test.

When I transposed these results with the loan loss provisions of three of the countries major banks we got an idea of the preparedness of Canadian financial institutions. My findings show that estimated losses were adequately provisioned for by all three banks (and in 2 of 3 instances, provisioned well above the 99% value at risk level). 2006 to 2008 saw an easing of the loan loss provision, but has subsequently been increased. Overall, I find that Canadian banks are well equipped to deal with even the worst of macroeconomic shocks. These results confirm what the International Monetary Fund found in 2007 when they assessed the stability of Canada’s financial industry and the World Economic Forums labelling of Canadian banks as the most stable in the world.

The inspiration for this paper came in early 2008, while the United States was experiencing a severe mortgage crisis. I wanted to see what effect macroeconomic shocks, similar to the situation in the US, would have on Canadian banks. In the year that followed the topic has become more and more relevant, with the credit crunch and subsequent collapse of banks around the world.

All data used is publicly available, mainly through Statistics Canada. My supervisor, Prof. Michael Bergman of the University of Copenhagen, provided some guidance mainly with regards to the model, but other then that all work has been done on my own without any special help.

i - Preface
ii – Table of Contents
1 – Introduction
2 – Related Studies
2.1 – Macroeconomic Effects on Credit Risk
2.2 – Stress Testing
3 – Theoretical Model
3.1 – A Macroeconomic Credit Risk Model
3.2 – Sectoral Exposure
3.3 – Loss Given Default
3.4 – Loan loss Distribution
4 – The Data
4.1 – Default Rates
4.2 – Macroeconomic Variable
4.3 – Sectoral Exposure
4.4 – Regional Exposure
5 – Empirical Model
6 – Stress Testing
6.1 – GDP Shock
6.2 – Real Interest Rate Shock- Canada
6.3 – Real Interest Rate Shock- USA
6.4 – Commodity Price Index Shock
6.5 – Dooms Day Shock
6.6 – Discussion
6.7 – Discussion About the Current Economic Crisis
7 – Implications, Limitations and Further Study
8 – Conclusion
9 – References
10 – Appendix A –The Data
11 – Appendix B – Additional Graphs

Stress testing is most simply defined as the process of determining the performance of an entity during extreme, yet not impossible situations. The test is used in a wide variety of disciplines from medicine to engineering and of course in the financial sector. The goal of the process is to identify potential vulnerabilities or to gauge the preparedness during extreme shocks. When applied to the financial sector, stress tests look at the robustness of a financial entity or instrument during a particular crash. The entity chosen can be anything from a financial institution to a securities portfolio. The crash or stress can come from anywhere, most commonly though are shocks introduced to the Gross Domestic Product (GDP), interest rates, unemployment, inflation or index prices (i.e. stock markets, commodity prices etc.) to name a few.

Stress testing in the financial sector has become widespread, with applications at both the micro and macroeconomic levels. Banks perform stress tests on their portfolios to ensure that reasonable allocations are made to cover potential losses. Governments and central banks perform stress tests to gauge financial stability on a national level. A further testament to the effectiveness of stress testing is its inclusion in the recent Basel II accord. The Basel II accord is a set of recommendations on banking laws and regulations issued by the Basel Committee on Banking Supervision(1). The accord seeks to ensure an adequate, risk adjusted determination of capital in all banks. The Bank of Canada, the nations central bank, supports the Basel II accord, and on November 1st 2007 implementation began at all Canadian chartered banks. In practice the financial system is too big and complex to apply a single stress test, so instead analysis is focused on a particular component, be it a portfolio or institution. In this paper, the focus is on the corporate loan portfolio of Canadian banks.

More specifically, this paper focuses on the effects of macroeconomic shocks on the credit risk of corporate loans in the Canadian banking industry. To do this I have created an aggregate corporate loan portfolio by combining proxy data for default rates of each industry in Canada. The default rates of the portfolio are used as a function of the macroeconomic environment, which are then subjected to stress scenarios. This paper follows the methods of Virolainen (2004) and Misina, Tessier and Dey (2006) (which adapt Wilson’s (1997) paper), which incorporate similar tests in Finland and Canada respectively. While Misina et al. also use Canadian data, they perform their test on a national scale where I have performed my test on a regional scale (i.e. for each province in Canada). The reason I have chosen to do so, is because the extreme size and diversity of Canada. I wanted to see if the stress tests could uncover particular regional weaknesses, while at the same time creating a more accurate corporate loan portfolio for Canada in aggregate. Pesola (2001), Vleighe (2001) and Boss (2002) have also contributed to the growing body of research on the macroeconomic determinants of credit risk within banks in their home nations, Scandinavia, Great Britain and Austria respectively.

According to the CIA World Factbook, Canada is an affluent, high-tech industrial society in the trillion-dollar class, making it a member of both the Organization for Economic Co-operation and Development (OECD) and the G-7. Overall, the economy is mainly serviced based; however due to the abundance of natural resources, Canada has a large commodity based primary sector. Exports account for roughly one-third of GDP, of which 80 percent goes to the United States. This has been facilitated by the 1994 signing of the North American Free Trade Agreement (NAFTA) between Canada, the US and Mexico. From a macroeconomic standpoint Canada is quite interesting because of it’s sheer size and the dispersion of its population. Behind only Russia in land area, Canada has 10 provinces and 3 territories each with a different economy. The service industry accounts for the largest percentage GDP in all provinces, however these are often based around a provinces historical industry specialization.

Starting in the east, Newfoundland and Labrador, Nova Scotia, New Brunswick and Prince Edward Island have long been reliant on the fisheries industry, though this has been in a steady decline because of overfishing. This has led to a shift towards mining, oil extraction and forestry. Meanwhile, Quebec has historically been focused on manufacturing, especially in the aerospace sector. Additionally, the province is the world’s largest producer of hydroelectric energy and is abundant with lumber and mining resources. Ontario is Canada’s financial center, where most corporations have their head offices (including the 5 big banks). The province is also the leader in manufacturing, namely in the automotive, iron and steel industries. Ontario actually surpassed Michigan (home of Detroit) in car production in 2004. Manitoba and Saskatchewan (nicknamed the Prairies) have long been associated with agriculture production in Canada though mining and petroleum now account for larger portions of the GDP. Alberta has become one of Canada’s strongest economic performers on the back of its substantial oil and gas production. Because most of the oil is trapped in oil sands, for a long time it was not a profitable industry. However, with the increase in oil prices and advancement of technology, the industry has since come to dominate the economy. The province is also an exporter of beef and wheat, which were chief industries before the oil boom. Finally, the most western province, British Columbia is dominated by the forestry industry but has also started to branch out into mining. The construction sector is also very strong in BC. The northern most part of Canada, including the Yukon, North West Territories and Nunavut, has long been considered desolate. Though, like in the rest of Canada, mining and oil exploration has created a number of business opportunities. Based on the wide variety of industries and resources throughout the nation, there is no question that macroeconomic shocks will affect each region differently.

Another interesting facet to the economy of Canada is the domination of the banking sector by five major banks. The Royal Bank of Canada (RBC), Toronto Dominion Bank (TD), Bank of Nova Scotia (Scotiabank), Canadian Imperial Bank of Commerce (CIBC) and the Bank of Montreal (BMO) make up over 85 percent of the Canadian banking market(2), with branches in every province (in addition to considerable international operations). While these banks have all diversified into wealth management and insurance, the retail division is still the largest. According to a World Economic Forum report, released on October 10th 2008, Canada has the soundest banking system in the world. The Financial Stability Assessment for Canada made by the International Monetary Fund (IMF) echoed this when it stated that “The five large banking groups that form the core of the system are conservatively managed and highly profitable. They have achieved strong risk-based capital ratios, modest returns on assets, and high returns on equity” and that “Canada’s institutional strength and robust framework for macroeconomic policies have underpinned solid growth and inflation performance” concluding that “Canadian banks appear to be sound and resilient”.

The inspiration for this paper came after the mortgage crisis hit the United States in the summer of 2007. I thought it would be interesting to see how Canada would react under similar circumstances, so in early 2008 I began research for this paper. What followed was a year of some of the greatest losses the real estate and stock market has ever seen. Now as the economies of the world are in recession, this paper seems all the more relevant. The mortgage and credit crises are discussed further in section 6.7.

This paper seeks to explore the stability of the banking sector in Canada. It is organized in the following way. Section 2 provides an overview of some related studies and the background behind the methodology. Section 3 develops the theoretical model. Section 4 introduces and describes the data being used. Section 5 then presents the empirical model. Section 6 presents the methodology of the stress test followed by the results. Section 7 presents the implications, limitations and possibilities for further study before finally presenting some concluding remarks in section 8.

1 - The Basel Committee on Banking Supervision is an institution created by central bankers from the G‐10 (which includes Canada) in 1974, which meets every 4 years with the goal of encouraging convergence toward common approaches and standards within the financial sector.

2 - This includes foreign bank subsidiaries.

Banks must consider many factors when judging credit risk, both micro and macro. It is intuitively clear that in economic downturns banks are more apprehensive about offering loans, especially to unproven customers. Conversely, banks are far more liberal in their lending habits when the economy is booming. The mortgage crisis in the United States (in 2007) and ensuing credit collapse is a perfect example of this. Banks that were willing to lend to anyone who asked became very cautious, refusing to lend to anyone, including other banks. These pro-cyclical tendencies lead to systematic instability as banks tighten credit when credit is most necessary (busts) and are quick to lend when central banks urge restraint (booms). Borio, Furfine and Lowe (2001) argue that this is caused by the poor measurement of changes in the absolute level of risk over time. This miss-measurement stems from the short horizons that underlie most risk measurement methodologies and from insufficient attention being paid to the correlations across borrowers and institutions. This means that when combined, changes in the risk associated with the economic cycle tend to be inaccurate. In particular, risk is often underestimated in booms and overestimated in recessions. The author’s claim that longer horizons and more focus on correlations would lead to better risk measurement. Allen and Saunders (2001) provide a detailed survey of the empirical evidence supporting the link between credit risk and macroeconomic developments. Their evidence indicates that macroeconomic conditions affect all aspects of credit loss, its probability, the amount recovered after default and the exposure at default. Virolainen (2004) points out that despite this correlation, most of today’s credit risk models do not explicitly links default probabilities and macroeconomic factors. Instead, most models include cyclical factors by applying different ratings transition matrices for “expansion”, “normal”, and “recession” period or by adjusting ratings transition matrices using a “credit cycle index”. While the cyclical effects of Merton-based market indicator models are captured primarily by the fluctuations in stock prices.

Crouhy, Galai and Mark (2000) provide a comparison of current credit risk models that have been developed by the banking industry. The KMV model, a Merton style model, has the macroeconomic exposure in the correlation of firms equity return. The CreditMetrics model has a transition matrix, which can be made dependent on the state of the economy. The new Basel II is a single factor model that assumes latent systemic risk factors. Recently, many of the bank models have developed a way to more explicitly link the impact of the macroeconomic environment.

While still not abundant, there is a growing body of research that investigates the macroeconomic determinants of credit risks within banks. Earlier work has focused on the loan loss provisions of banks and macroeconomic factors. Pesola (2001) examined the Nordic countries (Denmark, Sweden, Norway, and Finland) and found that corporate indebtedness had a significant positive effect on loan losses. While GDP growth (having a negative effect) and interest rates (having a positive effect) were less significant. Vlieghe (2001) modelled the relationship between the aggregate corporate sector default rate (in the UK) and macroeconomic factors. He found that property prices have a significant short run effect on company failures; while in the long run, he found that real interest rates and GDP are the significant determinants.

The above works tend to focus on current economic conditions and does not take into account worst case scenarios. Stress testing is a way for financial institutions to gauge their potential vulnerability to exceptional but plausible events (ie. a sharp drop in GDP growth). The tests have been used around the world by financial institutions since the early 1990’s, and are included in the Basel II capital adequacy framework. Sorge and Virolainen (2006) provide a comparative analysis of macroeconomic stress-testing methodologies. There are basically two approaches to stress testing, the Balance-Sheet and the Portfolio approaches. They both attempt to explain a set of indicators of financial position for an institution (or sector) in terms of some underlying variables.

The Balance-Sheet approach, as its name suggests, attempts to explain balance-sheet indicators in terms of macroeconomic variables. The indicators and variables used vary from model to model, depending on the goals of the paper. For example, Hoggarth, Sorensen and Zilcchino (2005) used a mix of macroeconomic variables (GDP, unemployment, house prices, commodity prices etc.) to explain bank write-offs in the UK. While Kalirai and Scheicher (2002) estimated loan loss provisions for the Austrian banking system using similar variables.

The Portfolio approach pools the assets (outstanding loans) of a bank or sector into a portfolio whose risk characteristics are then summarized in terms of a loss distribution. We then relate the loss distribution to macroeconomic variables. The effect of these variables can be seen in changes to the loss distribution. This approach is very similar to the Value-at-Risk approach used in microeconomic settings by individual institutions. The CreditPortfolioView (further developed by Wilson (1997)) was one of the first models to explicitly link the probability of default to values of future macroeconomic factors. These values are then used to create loss distributions. Boss (2002), Virolainen (2004), Misina, Tessier and Dey (2006) and Valentinyi-Endrész and Vásáry (2008) all use the portfolio approach, and are the base of this paper.

Boss (2002) was the first to apply a stress test to the Wilson (1997) model based on national banking figures. Boss’s model, designed for Austria, uses aggregate default rates and individual loan data for the exposure of banks. In order to get meaningful results from a stress test the scenarios must be extreme, yet not completely improbable. To ensure interesting results Boss (2002) used dramatic scenarios from Austria’s past (historically observed maximum changes in the macroeconomic indicators). He found that a general cyclical shock (i.e. a decline in industrial production), has the greatest impact on the loss distribution; with expected and unexpected losses increasing by more than 35% and 30%, respectively, (compared to the values determined on the basis of the scenario at year end-2001.) Oil price shocks provide the second largest impact, followed by disposable income, corporate investment and export shocks. Conversely, an interest rate shock, equity crash, or inflation shock, leads to a relatively small impact on losses. Generally speaking Boss finds that the Austrian banking sector’s risk-bearing capacity is sufficient.

Virolainen (2004) also uses the Wilson (1997) model, except as opposed to using aggregate default rates, he uses industry specific default rates for Finland; this allows for more accurate credit portfolio and loss estimates. Virolainen found that there is a strong connection between industry specific default rates and GDP, real interest rates and corporate indebtedness. However, the results are not uniform across all of the industries. Agriculture, for example, is only marginally affected when it comes to default rates and macroeconomic factors. In general, Virolainen (2004) concludes that credit risk stemming from GDP and interest rate shocks is modest for the Finnish banking sector.

Misina et al (2006) is closely related to that of Virolainen (2004) in that it uses the Wilson (1997) model and industry-specific bankruptcy rates as a proxy for default. The aggregate loan portfolio of Canadian banks is used as a function of the macroeconomic environment to conduct stress tests at the aggregate level. Their paper builds on Virolainen (2004) by using a vector-autoregressive (VAR) model as opposed to modeling the macroeconomic variables in isolation. This allows the effect of a change in one variable on the rest of the economy to be seen. The authors also added lags to help capture delayed impact of macroeconomic variables. Finally, they use industry level simulation to assess the impact on portfolio losses as opposed to individual company levels.

When stress testing their model, they chose to include select American macroeconomic indicators; this is because of Canada’s proximity and inter-dependence on the US. They found that, shocks to real interest rates had minimal impact on losses. However, shocks to the GDP of the US were much more severe, with expected losses being as high as 30 percent. In a combined, dooms-day scenario (an increase in commodity prices with a decrease in US GDP) the expected losses reach as high as 40 percent. The paper generally concludes that banks may be slightly under-provisioned. The authors are, however, quick to point out that the discrepancy between actual provisions and stress-induced losses may be the result of model deficiencies as opposed to problems with the actual provision. Though hindsight may point to the author’s being correct after all.

The final and most recent use of Wilson’s (1997) model comes from Valentinyi-Endrész and Vásáry (2008) using Hungarian data. In contrast to the Finish and Canadian authors, the authors chose to use a fitted autoregressive moving average (ARMA) model instead of the univariate autoregression (AR) or vector autoregression (VAR). They found that the most influential determinants of credit risk are the business cycle, the foreign interest rate and corporate indebtedness. The stress test results indicate that the Hungarian banking sector is robust and resilient to the relevant shocks.

This paper uses the same basic model originally developed by McKinsey and Co. (the Credit Portfolio View model). Wilson (1997) was the first to use the model to explicitly link macroeconomic factors and corporate sector. The model was also used by Virolainen (2004) and Misina et al (2006) in their studies of stress testing on corporate loan portfolios (as described in the previous section). The model is used to examine the link between default probabilities and macroeconomic factors. Once estimated, the model can be fitted to simulate the evolution of default rates in response to simulated economic shocks. Once we have estimated the future default rates we can make predictions about the future expected and unexpected losses for a credit portfolio, conditional on the current macroeconomic climate.

We will now examine Wilson’s (1997) model, as used by Virolainen (2004). Our goal is to model default probabilities as a function of macroeconomic variables. Wilson uses ratings to determine these probabilities, while Virolainen use the observed number of bankruptcies. The difference being that, ratings already include prompt macroeconomic developments, while observed bankruptcies may not. In this paper, I use the latter. In doing so we assume that all firms that file for bankruptcy have an outstanding bank loan. This is an overestimate, but without specific bank data it is the best we can do. In order to get meaningful results we must aggregate our observations of default rates. We do so by analyzing industry-specific default rates (we separate the bankruptcy rates by industry so that we can specify the sectoral exposure and create a more accurate corporate loan portfolio, this is discussed in section 4.3).

Virolarinen’s model consists of the following steps. The default rate for industry i is assumed to be a logistic function of an index y.

Where di,t is the default rate in sector i at time t. yi,t is a sector specific macroeconomic index, whose parameters must be estimated. We will model the macroeconomic index in such a way that a higher yi,t implies a better state of the economy, and therefore a lower default rate, di,t. The logistic functional form is used to ensure that default probabilities estimates fall in the range [0,1]. When the dependent variable is a probability (i.e. when using default rates), testing a linear relationship between explanatory and the dependent variable yields unreasonable results. To get around this, the preferred method is to apply a logit transformation. We obtain the macro index, y, by applying the inverse logistic function:

The industry specific macroeconomic index, yi,t ,is assumed to be determined by a number of key macroeconomic variables which influence the state of the economy. More specifically,

Where βi is a set of regression coefficients to be estimated for the ith industry (i= 1,2,….,n). xj,t is the set of explanatory macroeconomic factors (i.e. GDP growth in Canada and the US, inflation, interest rates etc.) at time t. νi,t is a random error (assumed to be independent and identically normally distributed) that captures the time and sector specific surprises. This model allows the sensitivity to different shocks to differ across industries (i.e. we can estimate different functions for different industries), while allowing the individual error terms to pick up the idiosyncratic shocks that are sector specific.

Until now we have specified a static model. The second step is to model and estimate the development of the individual macroeconomic time series describing the health of the economy. This is where Virolainen (2004), Misina et al (2006) and Valentinyi-Endrész and Vásáry (2008) all differ. The Finish and Hungarian authors use a form of the autoregressive processes (AR), the first using the equation of order 2 (AR(2)) across all variables. The latter fits an autoregressive moving average model (ARMA(p,q)) specifically for each variable. The Canadians (Misina et al), use a vector autoregressive model (VAR), arguing that it offers two channels of impact of a macroeconomic shock on default rates; the direct impact and the indirect impact from the effect of other macroeconomic variables. I choose to fit a ARMA(p,q) model, as it is close to the original Virolainen model while being better fitted, and is used in the most recent case. It is provided here:

where φj is a set of regression coefficients to be estimated for each macroeconomic variable, j and εt is a random error assumed to be independent and identically distributed.

Equations (1), (2) and (3) combine to form a system of equations which describes the

joint evolution of the industry-specific default rates and associated macroeconomic factors. A vector of error terms ((j+i) x 1), E, and a variance-covariance matrix of errors ((j+i) x (j+i)), Σ, are defined as follows:

It should be mentioned that the element of Σv sub-matrix could be highly significant, since bankruptcies in different sectors correlate not only because of their common dependence on systemic risk factors, as captured by equations (1), but also because of the direct connection between sectors, as found by Valentinyi-Endrész and Vásáry (2008) in their study of the Hungarian banking sector.

The final step is to use the parameter estimates, the error terms and the system of equations ((1), (2) & (3)) to simulate future paths of default rates across all industries over a specified time horizon. The forecasts of the bankruptcy rate can be made by using the Monte Carlo Simulations method. The simulation is carried out using the following steps:

(i) First, we generate a matrix of innovations by using the Cholesky decomposition (3). Simply put we create a matrix of normally distributed random numbers with mean 0 and standard deviation 1 and multiply it by the lower left Cholesky matrix (j x i) created from the covariance matrix of the residuals from all our variables. This gives us a random set of error terms with the same variance-covariance matrix as our model.

(ii) Using an initial set of values (say, mean of historical data) for the macroeconomic variables, xt, the parameter estimates (φj) of equation (3) and the generated macro innovation (ε), we calculate the forecast value for xt+1.

(iii) Using our forecast value of xt+1, the industry specific generated innovation, vi, and the parameter estimates ( b) of equation (2) we calculate the forecast for yi,t+1.

(iv) Using equation (1) and our forecast of yi,t+1 we calculate di,t+1.

We repeat steps (i) through (iv), using the previous forecast to generate the next periods. This is done a set number of times, depending on the desired time horizon. I choose to use an 8 period time horizon, or two years.

Figure 3.1 provides an example of what the forecasted time horizon might look like (using Ontario data). We can see that the industries are positively correlated, responding to generated innovations in a similar fashion.

3 - Cholesky decomposition is when a symmetric positive‐definite matrix can be decomposed into a lower triangular matrix and the transpose of the lower triangular matrix. Σ = AA’

We now have a forecast of default probabilities for each industry used in our model. Our next step is to combine these into a portfolio that accurately describes the make-up of the corporate loan portfolio of the banking industry.

The complexity of calculating the exposure for a portfolio depends on the type of credit instruments included. For example, if the portfolio is only made up of plain loans then the definition of exposure is the book value of the loans made to firms or, in an aggregate setting, the industry. The difficulty comes when off-balance sheet instruments are included. This is because it is not always explicitly clear where the credit risk exposure lies (for further reading see Saunders and Allen (2002)).

Virolainen (2004) creates his credit portfolio by using a database of financial statements for over 50,000 Finnish firms, almost 30,000 of which had outstanding loans. The credit portfolio is reduced to the 3000 firms with the largest outstanding loans, and then a 3% limit is put on any individual exposure in the portfolio. These firms are then divided into industries, where the values of their loans are added to give the value of the loans outstanding for each industry.

Misina et al. (2006) use aggregated non-mortgage loans issued for business purposes, reported in both domestic and foreign currencies. The authors point out that it would be best to use both domestic and foreign currency loans, but that it is far from straightforward. This is because one would have to trace the loans back to the each country and each industry and then model the default rates in the way described above, needing a multi-country model of macroeconomic variables. Additionally, the foreign exposure would be subject to variations in the foreign exchange rate, adding further complications. Due to the lack of available data, and the relative complexity the authors focus on domestic currency loans. This is the same method that I choose.

To get the value of the exposure for a particular industry you simply divide the loans made to that industry by the total of all loans. Details about Canada’s sectoral exposure are provided in the Data section (4.3).

The final factor in the model is to calculate the Loss Given Default (LGD), that is to say, the amount of money that is expected to be lost in the event of bankruptcy. It can be defined as LGD= 1- rr, where rr is the recovery rate (the amount of money recovered after default). To get the recovery rate for a particular industry we simply take the average recovery rate of all the loans in that industry.

Servigny and Renault (2004) point out that there are a number of problems when determining the recovery rate. The first, being exogenous factors; for example the quality of the collateral associated with the loan, where in line the bank sits when repayment begins or the ability to renegotiate the credit terms. Another problem is when to make the calculation, i.e. do you use the ultimate recovery (the amount the bank will eventually recover once the debt is settled) or the price of debt immediately after default. Both have drawbacks; ultimate recovery being long and difficult to assess when debt is not settled in cash while the immediate price of debt is likely to be underestimated, while the latter is susceptible to underestimation.

In practice, most models assume that the recovery rate is either constant or stochastically drawn from a particular distribution, and are uncorrelated with default. Altman, Resti and Sironi (2005) however, have found that recovery rates are not constant, but are in fact procyclical. Virolainen (2004) and Misina et al (2006) both assume recovery rates, and thus loss given default, to be constant. I will do the same, assuming a 50% recovery rate.

We are now ready to calculate the loan loss distribution for the corporate loan portfolio of the banking industry. We do this by combining the probability of default, the sectoral exposure and the loss given default.

Where llit is the expected loss on the loan, in industry i for period t; is the sectoral exposure of industry i, assumed to be constant; lgdi is the loss given default, for industry i, also assumed to be constant.

To generate a useable loss distribution we repeat steps (i) through (iv) and substitute the resulting default probability into (4). This process is repeated 5000 times to generate the loan loss distribution for an industry. All the industries are then added together to give the loan loss distribution for the corporate loan portfolio of the Canadian banking industry, by province. By splitting the data into provinces as well as industries, I am able to add another dimension to the loan loss distribution of the corporate loan portfolio. This is because of the size of Canada and the dispersion of its businesses. The regional exposure variable is created by dividing the value of non-mortgage loans in a province by the total amount of nonmortgage loans in Canada, giving a percentage estimate of a banks exposure to a particular province. Finally, I multiply the loan loss distribution of a province by its regional exposure, and sum the results for all provinces to create the loan loss distribution for the corporate loan portfolio in Canadian banks.

Note, that we can run the model for one province if we want to see the effect of a shock on specific regions in Canada. This is done by simply omitting the regional exposure variable and summing each industries loan losses as defined by equation (4).

Data on the number of bankruptcies per industry, per province is detailed and readily available. The data is available monthly from 1990, in a variety of industries and subindustries. These include: Agriculture, Manufacturing, Mining, Quarrying and Oil Drilling, Construction, Transportation and Communication, Utilities, Retail Trade, Wholesale Trade, Services, Accommodation and Food Services and Other. Bankruptcy data is obtained from the Office of the Superintendent of Bankruptcies of Canada (an agency of Industry Canada). The difficulty comes in choosing which industries to include when creating the corporate loan portfolio. Ideally, all would have a place, however, either because of too many periods with no defaults or because of inadequate exposure information some must be dropped. In deciding which industries to include, I focused on the frequency of bankruptcies during the sample period. The industries I choose are, Accommodation and food services (Accom), Construction (Cons), Manufacturing (Man), Services (Serv) and Retail trade (Retail).

This paper seeks to build on Misina et al (2006) by adding a geographic component to the examination of default rates within Canada. The reason for this being that Canada is the second largest country in the world in terms of area, but ranked 36th in population, making it amongst the worlds most sparsely populated countries. This has major effects on the industries in different regions. For example, Eastern Canada’s major industries include fisheries and forestry while Ontario is mainly services based and Alberta is oil rich. I have therefore, separated the data into provincial categories. Because of their size I have aggregated the smaller provinces in eastern Canada (Newfoundland, Prince Edward Island, Nova Scotia and New Brunswick are combined to create the Atlantic) and central Canada (Manitoba and Saskatchewan are combined to create the Prairies). I have also chosen to omit the Northern Territories (Yukon, North-West Territories and Nunavut) because there simply are too few establishments. This leaves us with 6 regions: The Atlantic (Atl), Quebec (Qb), Ontario (On), the Prairies (Pra), Alberta (Ab) and British Columbia (BC)(4).

4 - A full description of the data and where it was collected from can be seen in Appendix A.

To get default rates per industry we must use two different data sets, the number of bankruptcies per industry and total establishment counts per industry(5) (both are also separated by province). The fit is not perfect (because of how the industries are aggregated) but is accurate enough for our purposes. Unfortunately, the frequency of the data available does not match (monthly for bankruptcies, semi-annually for establishment counts) so to determine the quarterly bankruptcy rate we must:

• First, convert the monthly bankruptcy data to quarterly by summing each quarter’s monthly numbers.

• Then, we convert the semi-annual establishment counts be taking the average of the two adjacent semi-annual values to find the missing quarterly values.

• Finally, for each quarter we divide the total number of bankruptcies in an industry by the total establishment count for that industry.

Establishment counts are available by industry from 1997 to 2007 through Statistic’s Canada’s Canadian Business Patterns publication. Thus using the method above, we get quarterly data on default rates from 1997-q1 to 2007-q4 in 5 different industries and 6 different Canadian regions.

Figure 4.1 shows the bankruptcy rates for the five industries chosen, on a national scale. We can see that there has been a marked decrease in bankruptcies since the start of our sample period. Also of note is the apparent correlation between industries, indicating that most businesses are equally affected by macroeconomic shocks. Take for example the effect of September 11, 2001, all industries experienced a sharp increase in bankruptcies in the following quarter, a period marked by one of Canada’s only negative GDP growth rates during our sample period.

While bankruptcies appear to be correlated amongst industries, if we further subcategorize the data to include regional differences, we see something entirely different. Figure 4.2 shows the bankruptcy rates in the manufacturing sector of the different Canadian regions.

This graph indicates that while succinct during some major macroeconomic shocks (again we use periods following September 11th 2001, as an example), by in large the bankruptcy rates, in the manufacturing sector, among provinces appear to be uncorrelated. The manufacturing industry also appears to be the most volatile across the provinces. The construction sector appears to be more correlated (with the exception of Alberta, which is in the midst of a population boom), which may be because the longer time horizons faced by builders. The retail trade and accommodation and food services industries, appear to be somewhat in between the two, with respect to correlation, but are far less volatile across the Canadian regions. All industries do however exhibit the same downward trend to some degree. All graphs not included here can be found in appendix B.

Finally, we look at the bankruptcy rates of the five different industries within a province. Figure 4.3 shows the bankruptcy rates by industry within British Columbia. It appears that for most of the sample the industries are correlated, additionally there is a clear downward trend. This seems to be the case in Quebec and Ontario as well (these three provinces are biggest economies in the country), a fair amount of correlation and an easily observable downward trend. The smaller regions, including the Atlantic, Prairies and Alberta appear to be more volatile, with less correlation, but still a downward trend. This might be because these regions are smaller and more industry specific, increasing their sensitivity to shocks. Again, the graphs not provided here can be found in the appendix. I should point out that up until now my analysis has been largely superficial, based on graphs. This was ideal for observing trends but now I will focus on specific data, to more accurately describe what is happening in the various provinces and industries.

Table 4.1 provides a summary of the descriptive statistics for the bankruptcy rates for each industry and each province. We can see that the accommodation and food service industry exhibits the highest volatility, with standard deviation in a range between 0.001 and 0.0024, across the provinces. The service sector has the least amount of volatility with a range between 0.00024 and 0.00076. Amongst the provinces, Quebec has the greatest volatility, with a range of standard deviations between 0.00076 and 0.0024 across the five different industries. While British Columbia is the most stable, with a range between 0.00024 and 0.00064.

There is also evidence of asymmetry, which can generally be seen through the positive skewness coefficient. There are however some exceptions. For example, the construction industry is quite positively skewed, relative to the other industries. While the retail industry is much less so, and even negatively skewed in some provinces. The same is true amongst the provinces, with the Atlantic and Quebec being quite skewed, while the prairies are not.

The kurtosis varies greatly, both among industries and provinces. A high kurtosis value indicates that the variance is largely explained by infrequent, extreme deviations and vise versa. The construction industry has the highest peaks and is said to be leptokurtic, or a more acute peak and fatter tails (then a normal distribution). The retail trade industry exhibits cases of both leptokurtic and platykurtic distributions (or a smaller peak and thinner tails, then a normal distribution). Generally speaking, this means that the variance within the construction industry is more likely to be explained through large, infrequent deviations. The Atlantic region appears to have a sharper peak, while British Columbia has a distribution that is much flatter, meaning it’s variance is more likely described by frequent modest deviations. It should be noted that neither the provinces nor the industries have one easily observable kurtosis. The assumption of normality is rejected for all industries in all provinces, based on the Shapiro-Wilks test for normality at the 95% confidence interval. Meaning that the data is not normally distributed. Finally, we look at the correlation matrices of industries and regions, shown in Table 4.2 and Table 4.3 respectively. In both cases, there is no single way to describe the degree of correlation. When looking at the regional covariance matrices, we see that all industries in Quebec and to a lesser extent, Ontario and Alberta, are highly correlated. While in the Prairies, the manufacturing industry moves independently of the other variables. The same is true, albeit to a lesser extent, in the Atlantic region.

When we shift our focus to industries, and the correlation amongst regions (Table 4.3) we see that once again, it is the manufacturing sector that provides the weakest correlation. Quebec correlates to some degree with Ontario and Alberta as do the Prairies and Alberta, but not to the same extent as is observed in the other industries. Meanwhile, the service industry sees a high degree of positive correlation amongst regions. The construction, retail trade and accommodation and food services industries show a similarly high degree of correlation. I make the assumption that each firm within a Province or industry has an equal weighting, when creating the corporate loan portfolio. The problem with this is that bigger firms take bigger loans, and thus increase the exposure of the portfolio. One large firm defaulting could mean greater losses to the portfolio then say ten smaller firms. However, without data from every bank and every loan made, this is impossible to find. Virolainen (2004) did have access to such data and found that of the 58,000 firms included, the largest 3000 accounted for 94 percent of all outstanding corporate loans.

5 - Statistics Canada defines an establishment as “a statistical establishment is the production entity r smallest grouping of production entities which: a) produce a homogenous set of goods or services, b) does not cross provincial boundaries and c) provides data on the value of output together with the cost of principal intermediate inputs used along with the cost and quantities of labour resources used to produce the output.

The macroeconomic variables that we are most focused on are Canadian gross domestic product (GDP), real interest rates and the commodity price index. In addition to these, I look at the effect of the real interest rate from the US. All the data for both Canada and the United States is provided by Statistics Canada and is from the period 1997Q1 to 2007Q4, meaning there are 44 observations per variable.

The Canadian GDP data is seasonally adjusted at annual rates and chained in 2002 dollars. It is provided monthly, which I convert to quarters, by simply taking the month ending values for each quarter (i.e. q1 equals March’s value). GDP is converted into the GDP growth rate by subtracting the natural log of the current years GDP by the natural log of last year’s GDP (GDPg = (lnxt)-(lnxt-1)). The growth rate is then converted into a percent by multiplying by 100. Canada’s GDP growth rate is of importance for obvious reasons; if the economy suffers, we expect bankruptcies to increase. Figure 4.4 shows how GDP Growth behaved during our sample period. The data appears quite volatile, with some interesting macroeconomic shocks being quite visible; namely, the technology (or dot com) crash in 2000, the negative growth after 9/11 and the start of the mortgage crises towards the end of the sample. Therefore, I predict that an economic downturn in the Canada will negatively affect businesses, and thus bankruptcies. I would have liked to include the US GDP growth rate, but when applied to the model, it was simply not significant for any of the industries or regions. The GDP growth rates in Canada is however quite strongly correlated (0.53) with that of the US, so changes in US GDP can somewhat be seen through the Canadian GDP growth.

The nominal interest rate used is the Chartered Bank administered interest rate to prime business (or the rate offered to the most secure firms). The rates are taken from the last Wednesday of each month. This interest rate is basically the same as the Bank of Canada rate, with an additional 1.5% tacked on. I opt for the prime loan rate because it is the rate that firms would actually receive. The nominal interest rate is converted into real terms by subtracting the inflation rate (r=i-p)(6). The US real interest rates are also based on prime business rates minus inflation. Real interest rates are interesting because they affect the price of loans, and thus the ability of a firm to pay back a bank. I suspect that when real interest rates increase in Canada that defaults will also increase. The US real interest rate is of interest because it acts as a gauge for how the US economy is performing and since the North American Free Trade Agreement (NAFTA) has lead to over $1.5 billion in bilateral trade daily as well as 300,000 border crossings, US economic performance definitely affects Canadian performance. While US real interest rates do not directly impact the Canadian corporate loan portfolio, I suspect that if the rate decreases defaults will follow. This is because a comparative product in the US becomes less expensive to buy and make, thus decreasing exports. Therefore, I predict that an economic downturn in the US (leading to lower interest rates) will negatively affect Canadian businesses, and thus bankruptcies. Canadian and US real interest rates are highly correlated (0.85), with Canadian rates being higher during booms and lower during troughs. This is illustrated in Figure 4.5.

Finally, I have included the commodity price index, which is assumed to be determined in world markets (and is thus taken as exogenous). The data is in US dollars terms, indexed so 1982 to 1990 equals 100, which is then converted into the growth rate ( ). Commodities prices are very important; this is because when input prices go up, profit margins shrink, making business suffer. However, some areas in Canada are specialized in the extraction of commodities (for example, Alberta has oil, British Columbia, Quebec and Eastern Canada have forestry etc.). It will be interesting to see if some of these regions will be less affected by increases in commodity prices. I suspect, that the former will be a stronger influence.

6 - Inflation is calculated by using the Consumer Price Index as provided by Stats Canada with 2002=100 and this equation , US inflation use 1982‐84=100.

As described in section 3.2, to create an accurate corporate loan portfolio we must include the relative size of industries with respect to business loans, known as the sectoral exposure. I use quarterly data from Statistics Canada on non-mortgage loans made by chartered banks, by industrial classification (in Canadian dollars). By dividing the total loans made to an industry by the total of loans made to all businesses, we start to get a picture of the sectoral exposure in Canada (Figure 4.6).

During the stress-testing portion of the exercise we must decide on whether or not to use exposures from a specific date, or the mean. While there is clearly some fluctuation, all the figures are quite stable. Misina et al (2006) use actual exposures, so I will do the same. Both manufacturing and construction exposures appear to dip at the end of the sample while the service industry surges, but this is drawn out over several years, indicating a persistent trend. Thus I will use 2007q4 sectoral exposure values.

The final variable to consider is the regional exposure. The procedure for obtaining this statistic is the same as above. Using Bank of Canada data (via Statistics Canada), on the regional distribution of assets and liabilities for chartered banks (in Canadian dollars), we simply divide the total of all business loans in a province (or aggregated region, like the Atlantic) by the total in Canada. As with the sectoral exposure, there is little variance, but a couple of clear, trends emerge. Of most interest is the slow decline in Quebec loan activity, while Alberta has expanded. This has brought, British Columbia, Alberta and Quebec together, a stark difference when compared to the beginning of the period. Ontario still dominates, with almost half of all business loans in Canada going to its firms. This can be seen in figure 4.7. The decision to omit the northern region is proven to be sound, as it receives less then 0.01% of business loans provided by Canadian banks.

During the stress test, I will use the figures for the most recent period, 2007q4. This will provide the most up to date exposures and will match the procedure set up in the previous section.

We can now apply the data collected and described in section 4, to our model developed in section 3. The model is reiterated here for convenience:

Firstly, we must determine the set up of our macroeconomic index. This is done by determining the order of the autoregressive and moving average parts of each macroeconomic variable (i.e. the p’s and q’s of the ARMA equation (3)). I make the decision based on the Akaike Information Criterion (AIC), with up to 4 lags. The results are shown in Table 5.1, with the highlighted regions indicating the chosen ARMA combination.

Once the model is properly specified, I enter the estimated coefficients and the newly generated innovation term, as well as the chosen values for the loss given default (50%) and the sectoral exposures. This model is then simulated for 8 periods (2 years), and repeated 5000 times. The resulting distribution is the baseline, no-shock loan losses for the corporate loan portfolio of Canadian banks. During our simulation I must make some assumptions about the starting value of macroeconomic variables. There are two options; one can use the value of most recent period or one can take historical averages. The former is good for forecasting what will happen in the coming years, while the latter is better for determining losses during normal operating conditions. I have opted to use historical averages because the end of our sample period (2007q4) was uncharacteristically poor, with 2008 threatening to be even worse. By using historical averages, I can look at how Canadian banks are able to face shocks during normal operating conditions, as opposed to how they fair while the economy is already in recession.

The estimations of the effects of the macroeconomic variables on the various industries (for each region) are provided in Table 5.2. We can see that there are clear differences amongst both regions and industries. The real interest rates in Canada and the US have far greater impact in Eastern and Central Canada then in the West, where the variables are insignificant. The West is slightly more responsive to GDP growth changes then the East, but generally speaking is less affected by the set of macroeconomic variables, which can be seen in the lower R2’s. The R2 is in indicator of the goodness-of-fit for a model. In addition to indicating that the model is better fitted for the East then the West, we can see that the manufacturing industry is not very well fitted. In the cases where there is significance among the variables for the manufacturing sector, it is through commodity prices and interest rates.

When looking for the effects of particular macroeconomic variables on default rates, most are as I have predicted, but there are some surprises. As expected GDP growth has a negative effect on default rates, that is to say, as the economy grows, defaults decrease. The opposite is true for the real interest rate in Canada; as it increases so do defaults. This is as expected because when loans become more expensive, they are more difficult to pay back. Interestingly enough, as real interest rates in the US increase, defaults decrease. This is most likely because it decreases foreign competition. Also of interest is commodity prices, as they increase, so do default rates. Therefore, the prediction about input becoming more expensive, thus leading to more bankruptcies holds true.

Using the estimates from above I am ready to create the six regional loan loss distributions (later I will look at Canada as a whole, using my regional exposure estimates). The mean of each region’s no-shock loan loss distribution is provided (for each period) in Figure 5.1. The first thing we see is that Quebec has a much higher loss percentage then the rest of the country. This is explained by the higher then average occurrence of bankruptcies in the manufacturing sector (an industry with a high sectoral exposure). What is difficult to explain is why the default rates are so much higher in Quebec then in the rest of the country? The manufacturing sector in Quebec is smaller then in Ontario, in both total size and in terms of percent of total businesses. What is clear however is that banks should be aware of the increased risk of lending to Quebecois businesses compared to others. Generally speaking, western provinces lead to smaller percentage losses then eastern and central Canada.

Also of interest is the way some regions mimic each other in their path. For example the Atlantic, Ontario and Quebec move in unison, albeit at different levels. Meanwhile, the Prairies and British Columbia follow each other and Alberta is somewhere in the middle. The east coast (Atl), center (Qb & On) and west coast (Pra, Ab & BC) of Canada behave differently. Thus we can see that geography does seem to play a role in corporate loan losses.

The dashed, dark red line is the mean of the loan loss distributions for Canada, once the regional exposure is applied. The Canadian path follows that of Ontario quite closely; this is no surprise as Ontario receives almost half of all loans made to businesses in Canada. The higher loss percentage of Quebec counters the lower percentages exhibited in British Columbia and Alberta, resulting in Canadian loan loss percentages being slightly higher then Ontario’s.

7 - The missing values are because a flat log likelihood was encountered, so I (or rather Stata) could not find the uphill direction.

Now that the model is specified and estimated under normal operating conditions, we are ready to introduce shocks into the economy. Because of the diversity amongst the regions, I have to incorporate a variety of shocks, to see how each region react. The shocks I have decided to include are:

(i) GDP shock: negative 0.5% GDP growth for 4 consecutive periods

(ii) Real Interest Rate shock Canada: a 1.5 percent increase in the real interest rate in Canada, for 4 consecutive periods.

(iii) Real Interest Rate shock USA: a 2.25 percent decrease (increase) in the real interest rate in the US, for 4 consecutive periods.

(iv) Commodity Price shock: a 20% increase in the commodity price index, for 4 consecutive periods.

(v) Combination (dooms day scenario): the minimum GDP growth combined with the maximum real interest rate change in Canada for 4 consecutive periods.

The process for stress testing is the same as for the no-shock model. I enter the estimated coefficients along with an error (innovation) term, the loss given default and the sectoral exposure. The model is then simulated for 8 periods (2 years), and repeated 5000 times. The shock is introduced through the error term. This is done by simply inputting a decided upon value into the first period of the simulation (and subsequent periods if appropriate). The shocks introduced are meant to represent extreme, yet not unrealistic events, in an effort to gauge how the banking industry will fair during these incidents.

The simulated loss distributions for the corporate loan portfolios are presented in a couple of ways. Firstly, the summary statistics of the results are presented at the one-year and two-year time horizon. At both horizons the expected loss and unexpected loss are available. The unexpected loss is calculated by taking a set percentile of the results (I use the Value at Risk (VaR) at the 99% level), and then subtracting the expected loss (the mean). The non-stressed and stressed results are then compared. Also provided is the loan loss distribution of the corporate loan portfolio for Canada (after the regional exposure calculation has been inputted). The graphs show both the no stress and stress distributions, at both the one and two year time horizon(8).

I have attempted to maintain some uniformity across the stress tests. This is done by introducing the shock over the same number of periods (4, or one year) for each test. I have also chosen the new innovation term by taking the most extreme value experienced during the sample. The decision to use uniform tests, is so that comparison can be made more easily and accurately across the shocks.

8 - Because of the extremely long tails both the stress and no stress distribution, I removed the 99th percentile from the results (only when graphing). This allows more appropriate scaling, and more distinctive graphs. The 99th percentile, are reflected in the unexpected losses which are provided in the summary statistics.

Starting from the average rate of real GDP growth (over the sample), a stress of 0.5% negative growth, for 4 periods is introduced. Following the stress period, no form of shock or innovation is allowed. This is done so that the evolution of the shock can be seen over the simulation. The reason for maintaining the negative growth over several periods is to simulate a severe recession followed by a slow recovery. The cause of such an event is undetermined, but could be because of a negative export shock for example.

Results from the first year of the stress test, presented in Table 6.1.1, are quite strong. We can see that the results vary considerably between regions; they are nevertheless all significant. Alberta is by far the most affected by the GDP shock, with 126 percent greater expected losses then under normal circumstances. When looking at the unexpected losses, the effects follow suit, though in the Atlantic region the unexpected loss are considerably higher (the opposite is true for Ontario). This likely indicates a wider distribution, with a greater likelihood of big losses. Quebec and British Columbia are also quite affected by the shock, with expected losses of 60 and 74 percent respectively. This indicates that the distribution is flatter then average, and that there is a greater probability of larger losses. The Atlantic and Ontario are the most resilient regions, where a 2 percent negative GDP growth rate leads to a 28 percent increase in expected and unexpected losses. The Canadian portfolio saw a 56 percent increase in the expected losses, and an 89 percent increase in unexpected losses in response to the GDP stress scenario. We can conclude that the Canadian corporate credit portfolio would be quite vulnerable during a recession. It is important to also look at the actual percent of the credit portfolio that is expected to be lost. While Canadian expected losses increase by 56 percent, the actual percent of the portfolio lost is only 0.102 percent.

Figure 6.1.1 is a graphical representation of the loan loss distribution for the corporate loan portfolio in Canada. The no-stress distribution (the blue curve) is quite narrow around the mean, while the stress scenario shows less overall density and wider tails. The wider tails lead to greater unexpected losses, with more results at that extreme. Overall, we can see that the mean is dramatically shifted to the right, indicating greater losses.

Table 6.1.2 displays the results of the stress test after a period of 2 years. That is to say, after 4 quarters of negative GDP growth followed by 4 quarters without any innovations. The results are quite interesting, when compared to the results after year one. In all the regions, the stress tests show only a minor impact to both expected and unexpected losses. Alberta still has the largest percent change, but considering the scale of period one compared to period two’s (126 versus 4 percent), it is insignificant. The remarkable rebound is universal across all regions. Including Canada in aggregate with an expected loss increase of 2.31 percent and an unexpected loss increase of only 1.34 percent.

Figure 6.1.2 is the loan loss distribution of the corporate loan portfolio after the second year. Perhaps more apparent here then in the summary statistics is the fact that the economy is back to no stress loss levels. The stress curve is only a hair to the right of the original, with the tails overlapping as they disperse outward. Generally speaking, the effects of a GDP shock are severe while underway but recovery is very quick. This indicates that the corporate loan portfolio is not in long term jeopardy should the economy be hit by a recession.

The shock to the US real interest rate is a bit tricky to implement. This is because some provinces are positively affected by US interest rates, while others are negatively affected. The positive and negative values, coupled with the lower significances seen in Table 5.2 are indications of this discrepancy. Additionally, the US real interest rates appear to be more volatile then Canadian rates, with larger one or two period swings. For completeness, I have included both an increase and a decrease to the US real interest rate. Both represent a change of around 2.25% in four consecutive periods. I decided to use such a high value for an extended period because the most extreme cases were often followed by substantial changes in the same direction. I wanted to see the effect of one year of extreme interest rate changes, the likes of which the economy had not seen. An example of the conditions that could lead to this scenario might be that the nominal interest rate is decreased to spur investment but then deflation sets in, further lowering the real interest rate (and vice versa).

The first stress test initiated is a 2.25 percent increase in the real interest rate, for 4 consecutive periods. Table 6.3.1 shows how the effect of the increase impacts regions differently. Eastern and central Canada actually benefit from increases to the US real interest rate. This is most evident in Quebec, where expected losses decrease by 37 percent and unexpected losses fall by 44 percent. As we move west the effect switches, starting with the Prairies which sees no change what so ever. Alberta and British Columbia saw expected losses increase by 4 and 7 percent, respectively. These affects are minor when compared to the positive effect seen in eastern Canada (and the other stress tests). As mentioned in the Data section, a possible explanation for this might be because of the decrease in competition with US. As the cost of doing business in the US goes up because of domestic interest rates, Canadian products become more attractive, leading to more exports. The makeup of business in Canada, and the specialization of provinces might help to explain which side of the argument affects which region. Quebec and Ontario are mainly service and manufacturing based while the west is commodity based. So when interest rates go up in the US, American businesses import fewer input goods (oil, minerals, wood, wheat etc.), instead buying more finished goods (from central Canada).

Overall, Canada is positively affected by an increase in the American real interest rate. The impact however, is not very significant when we put the size of the shock into perspective. Over one year, interest rates are increased by 9 percent resulting in only a 22 percent decrease in expected losses. Unexpected losses generally follow suit. The loan loss distribution (Figure 6.31) is unconventional when compared to the other scenarios. Not in that it shows a decrease in loan losses, but in that it is more narrowly dispersed around the mean then the no stress situation. We can see that a US real interest rate is certainly beneficial, both in the decrease in expected losses and the reduction in uncertainty about losses.

In period 2 (Table 6.3.2) the results show that the effects of US interest rate shocks are longer lasting then in some of the other scenarios. All the results stabilize somewhat, but the decreases in expected losses are still quite high (19 percent in Quebec for example). Of interest are the Prairies, which initially saw no change but now has a positive effect compared to the pre-stress levels. Canada on aggregate sees the change in expected losses move closer to the no-stress levels, with only a minor decrease of 12 percent. Unexpected losses are slower to react, with the same percent change as the one-year interval. The loan loss distribution (Figure 6.3.2) confirms that the decrease in expected losses has subsided. The initial narrowing around the mean has also stopped, and now the curve is quite similar to the pre-stress curve. The loan loss distribution of the post stress curve demonstrates the slowed return to the status quo, especially when compared to the two previous scenarios. In this case, this is a good thing because expected losses are decreased, so a slower return to normal is desired. Intuitively this might be explained by business saving money during good operating conditions so they can survive longer in bad conditions. If the slower return to normal is also found in the US interest rate decrease scenario, this theory will be disproven (this however, is not the case).

Let us now implement an extreme decrease (2.25 percent in four consecutive periods) in the US real interest rate on our model. We expect the results to be the exact opposites of those presented above, which is almost the case, as seen in Table 6.3.3. As in the US real interest rate increase scenario, the impact of the shock changes as we move across the country. Eastern and central Canada see significant increases in the expected and unexpected losses. Ontario and the Atlantic actually experience the most severe changes to losses when compared to the other scenarios (not including the combination scenario); while the west coast regions see minor decreases. With a 57 percent change in expected losses, Quebec is the most affected by a decrease in the US real interest rate. Unexpected losses are all in line with their expected loss counterpart. Canada as a whole is quite negatively affected by a decrease to US rates, with 34 percent greater expected losses. This increase is in fact larger then that of a Canadian interest rate shock. This is dramatically seen in the loan loss distributions presented in Figure 6.3.3 where the curve is less densely distributed, has fatter tails and is pushed further right then under the Canadian interest rate scenario. However, when we compare the loan loss distribution of the US interest rate decrease to a decline in GDP growth, we see that the latter is far more serious. The GDP distribution would not even fit on the same scale. Nevertheless, it is safe to say that the US (namely the interest rate) plays a pivotal role on loan losses to the Canadian corporate loan portfolio.

When we look at the results after the second year (Table 6.3.4), we see that, once again, increases to expected losses drop dramatically. Quebec still has the largest percent change, but with that number falling by 75 percent we can safely say that the province rebounds quickly. The same is true for the other eastern regions, which are negatively affected by the shock. In western Canada, which is positively affected by US interest rate decreases, There is a movement towards normal conditions as well. Canada as a whole proves to be quite resilient, with expected loss increases falling from 34 percent to 10 percent over the base year. This movement is very similar to that of the Canadian interest rate scenario. In both instances the curves have the same characteristics as the no stress distributions, while being moved slightly to the right.

When the US interest rate decrease distribution is compared to the US interest rate increase distribution, we see that the former is much closer to the no stress distribution. This lends evidence to the theory that the affects of positive shocks to the corporate loan portfolio dissipate slower then the effects of negative shocks.

The next stress test performed is on the real interest rate in Canada. I follow a similar approach to that of the GDP stress test, in that I take the most extreme interest rate change during the sample, 1.5 percent, and input it into the first 4 periods of the sample. This simulates an extreme interest rate increase for a one-year period. This might occur if there is persistent deflation without nominal rate changes. Nevertheless, this will demonstrate the bank sectors resilience if such an event occurs.

The results from the stress test, which are presented in Table 6.2.1, once again show significant variation amongst the provinces. Overall, the impact of the interest rate shock is far weaker then that of the GDP shock scenario. The regional effects of this stress test are interesting because they show a clear difference between the East and West coast of Canada. The Atlantic region is the most effected under this scenario with an expected loss change of 44 percent. Similar results are seen in Quebec and Ontario. Meanwhile, British Columbia exhibits basically no change considering the size of the shock (a 6 percent increase in interest rates leads to only a 3.4 percent change in expected losses). The same is true in the Prairies. Alberta has a slightly higher percent change of 20 percent, but considering that it was the most effected region by the GDP shock, I believe the notion that the west is less susceptible to Canadian interest rate shocks holds. The increases to unexpected losses are for the most part smaller then the expected losses. Up until this point, it appears that British Columbia and the Prairies are the best place to offer credit. This might be somewhat misleading because of both regions focus on agriculture (not included here), which is much greater there then anywhere else in Canada.

When we aggregate the model using the regional exposure estimates, we get an indication of how the Canadian credit portfolio fares under this extreme circumstance. With expected and unexpected losses both hovering around 27 percent greater then the no stress simulation, the loan portfolio appears vulnerable to prolonged, extreme interest rates. This intuitively makes sense as interest rates go up, loans become more expensive for businesses and thus more difficult to pay back, leading to losses for the banks loan portfolios. The loan loss distribution, presented in Figure 6.2.1, helps to show that losses are not as severe as under the GDP scenario. Additionally, the stress curve has almost the exact same characteristics as the no stress test curve, but has simply been moved to the right.

When we shift our focus to a longer time horizon, we see that the results are about the same. Table 6.2.2 displays the findings of the interest rate stress test after two years. We can see that the percent increase in expected losses are slightly higher then under GDP stress scenarios. Again, it is the east coast regions, the Atlantic, Quebec and Ontario, which have the highest percent changes, while the Prairies and British Columbia have almost not change. The unexpected losses in this example are quite interesting in that they are almost all negative. Canada as a whole, recovered quite well in year two with an expected loss increase of 7 percent, this is about a 70 percent decrease compared to the first year’s results. The unexpected loss was lower after the shock then before, with a percent decrease of 1.9. The speed of recovery is a bit more sluggish then after a GDP shock, but still quick overall.

The stressed loan loss distribution after the second year, as in the first, has the same characteristics as the no stress distribution. The graph (Figure 6.2.2) does indicate that losses are greater then when there are not persistent interest rate increases. While this affect is not overwhelming, it is greater then the affect on the loan loss distribution after the GDP scenario. We can summarize by saying that a high interest rate over several periods has a significant effect on Canadian credit portfolios, responds relatively quickly when the shock is removed, but some increased losses persist.

The final single variable shock I examine is a persistent increase to the commodity price index. The value decided upon is a 20 percent increase, implemented in 4 consecutive periods. Commodity prices are determined on the world market, and affect the cost of doing business everywhere. While Canada is quite a large commodities exporter (particularly oil)(9), Ontario and Quebec, the two largest receivers of corporate credit, are mainly manufacturing and service based. The shock therefore could impact Canadian regions in two ways; either by lowering world demand of commodities (causing Alberta to suffer) or by increasing the cost of production (adversely affecting Ontario and Quebec). Whatever the effect, I expect that this shock will have a quite a large negative effect on losses. The results, presented in Table 6.4.1, confirm the negativity of the shock, however the sizes of these losses are not very large compared to the other scenarios.

Unlike the other stress tests, the increase in commodity prices affects expected losses uniformly (for the most part) across Canada. Alberta once again sees the largest percent increase in expected and unexpected losses, with a value of 39 percent, almost double that of the next largest region affected, the Atlantic. Ontario is the least impacted province with an increase of 14 percent. Again in contrast to the other stress tests, unexpected losses are greater in every instance. This indicates that the loss distribution is more widely dispersed around the mean then in the other examples. (this is somewhat confirmed in Figure 3.4.1) Based on the considerable difference between Alberta and every other region, we can assume that changes in commodity prices have a greater impact on demand compared to the input cost effect. When the regional exposure is accounted for, Canada experiences a 20 percent increase in expected losses and a 50 percent increase in unexpected losses. This is insignificant when considering the scale of the shock, an 80 percent price change in just one year. This is therefore, the least severe of all the stress test scenarios. The loan loss distribution, presented in Figure 3.4.1, shows that the curve is fatter, and that the tails are much longer compared to the no stress distribution. Other then the relatively large tails, this shock is not as interesting compared to the others already performed, and does not pose a significant threat to Canadian.

At the two-year time horizon, presented in Table 6.4.2, the results are slightly different then we have come to expect. While expected losses have all decreased considerably, a couple of the unexpected losses remain quite high. British Columbia for example now has a higher increase in unexpected losses then after year-one (48 percent versus 40 percent). The fact that expected losses rebound but unexpected losses do not, indicates that after the stress test, the loss distribution is far more dispersed resulting in a greater probability of extreme losses. This effect is severely dampened when the results are aggregated to make the Canadian loan portfolio. Expected losses fall to 6 percent while unexpected losses fall to 14 percent, a 70 percent difference over the first period. The resulting loan loss distribution (Figure 6.4.2) is for the most part identical to the no stress distribution, with the exception of a slightly longer right tail. This is the effect of the large unexpected loss. These results solidify the conclusions drawn in the previous paragraph, that the Canadian corporate loan portfolio is resilient towards commodity price shocks, even when severe.

9 - Canada has the second largest estimated oil reserves in the world, behind only Saudi Arabia. Of which, 95% is located in Alberta.

We have now seen how the corporate credit portfolio reacts to various stress scenarios, but what would happen if more then one shock occurred at once. To create the Dooms Day scenario, I include an extreme GDP growth rate decrease and an extreme Canadian real interest rate increase. This will explain what would happen if the economy moves into a recession while interest rates continue to climb. While this is rare, it is not unheard of in history. The Great Depression is a perfect example, where a deflationary spiral(10) meant real interest rates continued to climb, while the GDP fell drastically. During this stress test, the GDP growth rate will decrease by 0.5 percent and the real interest rate will increase by 1.5 percent for 4 consecutive quarters. I did not include shocks to the US real interest rate or commodity prices because they are both exogenous variables (i.e. uncontrollable through government policy). Table 6.5.1 provides the results after one year.

As expected the results are quite dramatic. It is no surprise that Alberta is the most affected considering the results from the GDP stress scenario. With an increase in expected losses of 169 percent, it is fair to say that the corporate loan portfolio is highly vulnerable in Alberta under the dooms-day scenario. There is no real pattern (as in the interest rate scenarios) to the way expected losses are affected throughout the regions. Ontario, British Columbia, the Atlantic and Quebec all experience similar changes, with losses increasing in the range of 71 to 103 percent. The Prairies are the most resilient to this scenario with expected losses only reaching the 49 percent level. Unexpected losses show smaller increases, but follow the same pattern as expected losses. Canada experiences a 95 percent increase in expected losses and an 80 percent increase to unexpected losses. These are, not surprisingly, the biggest losses experienced during the stress test exercise. What is interesting is that the losses under this scenario are not simply the sum of the GDP shock losses and the real Canadian interest rate shocks (which would be about an 83 percent increase). This means that the two shock combine to create losses that would not have occurred under the individual scenarios (i.e. a business might be able to survive high interest rates or a slowing economy, but not both). The corporate loan loss distribution (Figure 6.5.1) shows just how dramatic these losses are. The two curves barely fit on the same graph. The no-stress curve is tall and thin, while the stressed distribution is widely dispersed with very long tails.

Shockingly, by the end of year two, the expected losses have almost returned to their pre-stress levels. This is seen in table 6.5.2. While most of the regions still have expected losses with around a 10 percent increase, considering the levels after year one, these are insignificant. Some regions, including Canada as a whole, show decreases in the change of unexpected losses. Figure 6.5.2 helps to visualize the change. The loan loss distribution has lost all the characteristics displayed in the year-one graph. The two curves are basically identical, with the stress distributions shifted right, with its tail converging on the non-stress distribution. This indicates that no matter the size of the shock, given 4 periods without any innovations the loan loss portfolios will come close to returning to their original loss distributions.

10 - A deflationary spiral occurs when decreases in prices leads to lower production, which in turn leads to lower wages and demand, which leads to further decreases in price‐ and thus deflation.

Overall, the stress test results behave in the way we predicted. A shock to the real GDP growth rate has the largest impact on the corporate credit portfolio, both on a regional and national scale. An extreme increase in the Canadian real interest rate was less damaging then I would have assumed, but still behaved as predicted (by increasing expected and unexpected losses). US real interest rates were a bit more finicky, causing increased losses in some regions, while decreases in others. On the national level, a decrease to the rate led to substantial changes in the expected and unexpected losses. The corporate loan portfolio proved to be most resilient towards commodity prices across the country, with a substantial increase in prices leading to a relatively small increase in losses. And finally, a combination scenario led to sharp increases in the expected losses, greater then simply the sum of the GDP and real interest rate scenarios, hence its nickname being the Dooms Day scenario. In almost all instances the effects of the stress test dissipated by the end of the second year.

This is clearly seen in Figure 6.6.1, which maps the evolution of expected losses to the Canadian credit portfolio after each stress test. What we see is that after the 4 periods of stress it takes about 2 quarters for losses to dissipate. The increase in losses never fully disappears, but the difference falls to less then 0.008 percent (at its greatest). The GDP stress finishes closest to normal, followed by the real interest rates and the combination scenario. Commodity prices and GDP growth in period 3 are the only instances of divergence from the general direction of the no stress line. We can also see that the GDP and real interest rate shocks introduced are all about the same severity, while the commodity price shock is almost twice as large relative to the starting point. This lends further credence to the conclusion that commodity price shocks are the least significant. These results are somewhat different when we divide the country into region.

The Atlantic region, made up of Newfoundland and Labrador, Nova Scotia, New Brunswick and Prince Edward Island is the countries smallest economy. The region is mainly agriculture based, centered around its large fishery industry. Though some oil extraction has begun, as well as other diversification because of the falling fisheries industry. It is most greatly affected by a decrease to the US real interest rate. The region is also highly impacted by an increase to the Canadian real interest rate. While the GDP is still a factor, it is far less severe then in the other areas. The combination scenario is quite severe, with the third largest increases to expected losses in the nation. We can conclude that the corporate loan portfolios are quite safe in the region, as long as a combination of GDP decreases and real interest rate increases do not occur.

Quebec is Canada’s second largest economy, with strong manufacturing, services and exports sectors. The corporate loan loss portfolio of the region is most susceptible to decreases in the GDP growth rate. The US real interest rate is also a sensitive area, which is likely explained by the reliance on exports going to the US. The Canadian real interest rate shock created substantially greater losses, though the commodity price change did not. Quebec was seriously impacted by the combination scenario, with over 100 percent greater losses, the second greatest increase in the nation. The fact that the region was quite susceptible to four of the five stress tests makes it apparent that Quebec has the most fragile corporate loan portfolio in the nation.

Ontario is Canada’s largest economy, with a GDP nearly twice the size of the next biggest province, Quebec. It is for the most part service based, but also has a large manufacturing industry and under exploited stash of natural resources. The stress tests reveal that Ontario is most affected by decreases in the US real interest rate. Canadian real interest rates also create large loss increases, while commodity prices do not. The GDP shock does increase the expected losses but compared to the other regions, the degree is not very large. As is the case in all provinces, the combination scenario is quite severe, but comparatively speaking Ontario’s credit loss portfolio is resilient. It is the second-least affected by the extreme scenario. Overall, I consider Ontario’s corporate loan loss portfolio to be very resilient towards unexpected shocks.

The Prairies, made up of Manitoba and Saskatchewan, is Canada’s second smallest region (though it receives the least amount of corporate loans). It is agriculture based with substantial exports around the world (giving it the nickname the bread basket of the world). The model does not include agriculture as a sector because of its small size in the larger economies, so these results must be taken with a grain of salt. The Prairies are most affected by decreases to the GDP growth rate. In fact, none of the other stress tests have much impact on the corporate loan portfolio at all. Commodity price shocks do increase expected losses by 15 percent, but as mention in section 6.4, relative to the size of the shock the value is insignificant. The region also reacts best to the combination scenario with expected losses increasing by only 49 percent. This shows that the Prairies is the safest region when comparing corporate loan portfolios11.

Alberta is now the third largest economy in Canada, recently over taking British Columbia on the back of its burgeoning oil industry. The boom has led to a rapidly increasing construction industry, as well as an influx of service-based businesses. Alberta differs considerably compared to central Canada, namely in the positive effect of decreases to US real interest rates. It is however very susceptible to other shocks. Alberta is far and away the most affected by the GDP stress scenario, with a 126 percent increase in expected losses. It is also the only region to experience a significant effect from the commodity price increase scenario. The Canadian real interest rate stress test has only a small impact compared to the central and eastern provinces, but was larger then its western counterparts. Finally, the combination scenario shows the most extreme increases to expected losses in the nation. Overall, the regions corporate loan portfolio is very fragile, but only when the economy as a whole is in decline.

The forth largest and most westerly, province in Canada is British Columbia. The region is largely resource based (centered around forestry) though the construction and service industries are also thriving. The trend in the western Canada is reinforced by British Columbia in that it is most affected by a GDP shock with the second most severe increase in expected losses nationally. The rest of the stress scenarios are however ineffective. The real Canadian interest rate change leads to only single digit increases, while the US rate decreases actually lead to smaller expected losses. As in the other regions, commodity prices are insignificant. British Columbia falls in the middle when looking at combination scenario, with its value almost complete explained by the GDP growth rate increase. This leads to the conclusion that the region is quite resilient, but given a big enough GDP decrease, the loan loss portfolio will suffer.

Thus far all my conclusions on the stress tests have been based on the percent increase of the expected losses, but is this a fair assessment? Considering expected losses never move higher then 0.15 percent of the portfolio it appears the banks are fairly safe, but before making any firm conclusions we should look at the historical loan loss provisions in Canadian banks. The loan loss provision is the amount of money the bank has set aside to cover losses from defaulting loans. Figure 6.6.2 shows the loan loss provisions of 3 of Canada’s 5 major banks (12).

The graph also shows the expected losses of our simulation without stress and the most extreme losses under stress (the 99th percentile losses of the combination scenario)(13). We can see that all 3 banks are above the no stress line, meaning that all the banks have provisions enough for the expected loan losses of their corporate credit portfolio, under normal operating conditions. The more interesting story is that of the stress region. All three banks start out in between the stress and no stress line; this indicates that they are at least somewhat susceptible to the increased losses generated in our tests. The Royal Bank of Canada (RBC) moves above the worst-case scenario after just one year. The high loan loss provision persists from 1999 to 2003, a period which includes the tech crash and 9/11, but then declines (past the worst-case scenario line) until late 2007. Toronto-Dominion bank (TD) takes longer to increase its loan loss provision above the stress region, but after doing so does not let it fall below. The Bank of Montreal (BMO) is the obvious outlier, consistently living in the stress region (up until the second quarter of 2008). This means that BMO is quite susceptible to the stress tests introduced in earlier sections and should consider raising its loan loss provision. These results and conclusions should be taken with a grain of salt, as the banks have far more accurate data and test available to them. So generally speaking, Canadian banks are quite prudent and forward thinking, which leads to the conclusion that the industry is well prepared for even the worst of economic conditions

11 - If we do not include the agriculture industry, which accounts for 1/3 of all businesses in the region.

12 - The Royal Bank of Canada (RBC) the countries largest bank by market capitalization, Toronto Dominion Bank (TD), the second largest and the Bank of Montreal (BMO) the smallest of the ‘Big 5’.

13 - The most extreme losses are used because then all other losses under stress fall between that line and the no stress line, so if the loan loss provision is above this line, it is able to weather any form of shock.

The origins and causes of the current crisis are many but generally speaking can be explained by deregulation, speculation and the lack of understand of derivatives. After the technology crash, interest rates were very low, this forced financial institutions looking to invest (which because of deregulation now included both commercial and investment banks) to seek new, profitable investments. What they found were collateralized debt obligation’s (CDO’s), which in some (but not all) cases were created by pooling the mortgages of a bank. There was a constant stream of revenue (people paying their mortgages), with little risk (default rates where low, and by pooling meant any defaults risk was marginalized). CDO’s became very popular, which led to more demand for mortgages. This led to sub-prime (and in some cases predatory) lending, which is the practice of giving loans to people who normally would not qualify, either because of a bad credit history or an inadequate income etc. As house prices continued to increase and interest rates were relatively low, things were good. However, interest rates increased and more and more of these sub-prime mortgages began to default, leading to a fall in real estate prices, more losses and the eventual crash in the mortgage market. This was disastrous for banks that had large accumulations of CDO’s, filled with these toxic loans. This led to huge write-downs (14) among banks around the world, and eventually the failures (and in some cases bankruptcy) of some of the world’s biggest financial institutions. Despite bailout attempts by the US government, banks became weary of lending to each other and credit eventually seized up.

This has had global ramifications. Financial institutions around the world, which had accumulated CDO’s where directly impacted and forced to accept big write-downs. The effects have also been indirect. Because of shattered consumer confidence and spending in the US, stock markets around the world have nose dived. Liquidity problems have cropped up in both developed and emerging economies, forcing the IMF to offer emergency founds in a number of nations. The US is confirmed to be in recession, while growth has slowed around the world.

The impact of all of this on Canada is obvious. Firstly, the US is Canada’s largest trading partner accounting for some 80 percent of all exports. A recession in the US means many Canadian businesses are adversely affected. As stated earlier Canadian banks are far more focused on the commercial aspect of banking, but some write-downs did occur. Also, almost all of the Big 5 Canadian banks have international operations in the US. Though those businesses might suffer, some argue that there are good opportunities for expansion in the struggling US market. Growth has slowed considerably, but is still positive, with some key industries affected (manufacturing for example). The government is apprehensive about deficit spending (the budget has been balanced for over 10 years) to stimulate the economy, but it appears inevitable.

This turmoil is very interesting (if a bit frightening) from an economic standpoint, however is not included in my model. The data I use is only up until the fourth quarter of 2007. The reasoning for this is two-fold; in practice, the data on bankruptcies rates and GDP is simply not available. Even if it were I might not include it for fear that it would distort my results. This event is not business as usual, and thus any stress would have to exceed it, creating improbable circumstances. While it might be interesting to see what would happen to the banking sector should a stress occur while already in a poor financial state, this paper does not explore that option. What this paper does provide is insight into how the banking sector is equipped to handle this crisis.

14 - Reducing the book value of an asset because it is overvalued compared to the market value.

The implications and limitations of this paper are quite intertwined, as is the progression of study using this model. While this paper follows the method established by central banks around the world, there are some issues that need to be addressed. Namely, the assumptions made about the available data and its effectiveness on our model. The first and most glaring assumption made is that bankruptcy rates are a good proxy for corporate loan defaults. While it is safe to assume a company that defaults on a corporate loan is in the throws of bankruptcy, we cannot explicitly say that all bankrupt companies default on loans (or even have bank loans). Companies have a variety of methods for gathering funds that do not include banks, stock offerings (in large corporations) or borrowing from family and friends (in small businesses) for example. Also not included are the funds borrowed from outside of Canada. Therefore, in assuming every company in Canada borrows from a Canadian bank our credit portfolio becomes distorted. Additionally, my methodology does not consider the amount of credit offered to each business. The effect of bankruptcies in multi-billion dollar oil firms and a mom and pop shoe repair stores are thus given the same weight in this model. Even if we consider bankruptcy rates a good proxy for default rates, not all Canadian industries are included, agriculture being a notable example. Therefore even with perfect data for the included industries the corporate credit portfolio is incomplete. And finally, by assuming that the loss given default is the same for every industry, in every period is unrealistic, as described in section 3.4. These assumptions were unfortunately necessary based on the available public data. Banks, with actual data on the makeup of their corporate loan portfolio, default rates and recovery rates are able to create far more accurate models, as I’m sure they do.

Despite the shortcomings in the model and data, I believe it is fairly well developed when looking at the aggregate picture of the Canadian financial industry. By applying the stress tests we can see if there are specific vulnerabilities within the banking industry. In this paper I divided Canada into regions, to get a sense of how the provinces react to the shocks introduced. The portfolio can be even further subdivided by looking at specific industries, on both the national and regional scale. In doing this, we can get an even better impression of the soundness of the industry and if there are industries and/or regions that are particularly vulnerable to different shocks. Additionally, the stress test helps to indicate the relative effect of the macroeconomic variables introduced. For example, my results show that increases in the commodity price index has a relatively small impact on expected losses. By introducing more variables, both in aggregate and on specific regions and industries, we can develop a greater understanding of how the variables affect credit risk. This leads to a number of possible policy implications both at the government and firm level.

On a national scale, by running stress tests on the banking industry, the government can see if possible regulations or intervention is necessary. For example minimum loan loss provisions could be implemented, though this is unlikely and is not included under the Basel II accord. More likely is that during stress situation that are identified as of serious concern, the Bank of Canada could increase the access to liquidity. The result of this being that, should a number of corporate loans default, the chartered banks could borrow cheaply from the Bank of Canada, allowing them to whether the storm. By identifying which macroeconomic events and in which regions, or sectors are most susceptible, the national bank can react more aggressively to help the chartered banks. The government could also take a proactive stance. For example, if a region is identified as particularly susceptible to a shock the government can introduce tax breaks or other stimulus in the region to decrease the likelihood of defaulting. The same can be done if a particular industry is identified as fragile. Additionally, if a certain macroeconomic event is shown to be particularly hazardous, the government or national bank (depending on the shock) can target it more or less aggressively as it sees fit. For example, if Canadian real interest rates had proven to more damaging to expected losses, the national bank could make sure that it did not increase it too much too quickly. Fortunately as described in section 1 Canadian banks, and the financial system in general is very well run, and does not need much government intervention.

At the firm level these results are less interesting, because of the available data and the state of the art models that banks use. However, when running similar tests, the banks might identify industries or regions in which more caution should be exercised. This would allow more accurately set loan loss provisions or higher interest rates for problem areas (i.e. a risk premium). It could also lead to more scrutiny of possible obligors in some industries while a more relaxed approach in others, resulting in a more efficient use of resources.

There is a great deal of possible future research in this area, using the basics of this model. As mentioned, greater subdivision of regions (from provinces to cities etc.) and/or industries will lead to more specific results. The macroeconomic variables could also be made more specific, for example instead of using Canadian GDP growth, one could use specific regional GDP’s. It would also be interesting to include the effects that shocks on one variable have on others. For example, Canadian and US real interest rate are clearly correlated, so if there is a shock in one, there should also be changes in the other (which is not consider in my model, but is in Misina et al (2006)). Another possible simulation would be to use firm level shocks, as in Virolainen (2004), where instead of using industry level default probabilities he generates default events at the firm level. This would help to reduce the effect of defaults in small businesses versus big business. Finally, this paper only uses non-mortgage loans (i.e. business loans), an interesting application would be to include both (though then information on the amount of defaults and size of loans is scarce). This would provide a more accurate view of the total loan losses experienced by Canadian banks and not just corporate loan losses.

With world economic conditions being what they are, an investigation into the impact of shocks on the credit risk of banks is quite timely. In this paper, I have estimated a macroeconomic credit risk model for the corporate loan portfolio of Canadian banks in the nations six relevant regions. To create the corporate loan portfolio I aggregated the default rate of a number of industries based on their sectoral exposure. In using data from the last ten years, the sample includes both booms and busts of the economy.

The results from the empirical model show some variance in the significance of the relationship between the bankruptcy rates and the macroeconomic variables chosen. GDP and commodity prices are significant in almost all instances, while real interest rates are a bit more finicky, especially on the regional level.

When stress is applied to the model, the results are as we might expect. A severe decrease to the GDP growth rate negatively impacts default rates in all industries in all regions. The same is true for the real interest rate, all be it to less of a degree. Commodity price shocks prove to have little impact on default rates, while US real interest rates affect regions differently (in aggregate a decrease leads to greater expected losses).

The regions prove to be affected at different levels by the shock scenarios. Quebec is the most susceptible to all the shocks, while Alberta sees the greatest expected loss increases, though is not affect by real interest rate changes. The Prairies are the most resilient towards the shock, but that might be explained by the exclusion of agriculture in the corporate loan portfolio.

The model developed in this paper may be inadequate on a firm bases (because of assumptions made and the limits to data), it does however provide a pretty good picture of the stability of banks on a national scale. Overall, Canadian banks appear to be adequately prepared to face the extreme stress scenarios created in this paper; with loan loss provisions of the biggest Canadian banks well above even the most extreme losses. These results confirm what the International Monetary Fund found in 2007 when they assessed the stability of Canada’s financial industry and the World Economic Forums labeling of Canadian banks as the most stable in the world.

Allen, L. and Saunders, A. (2003). A Survey of Cyclical Effects in Credit Risk Measurement Models. BIS Working Papers, No. 126, January 2003.

Altman, E., A. Resti, and A. Sironi. (2004). Default Recovery Rates in Credit Risk Modeling: A Review of the Literature and Empirical Evidence. Economic Note, No. 33, 2004.

Blaschke, W. M. Jones, G. Majnoni and S. M. Peria. (1988). Stress Testing of Financial Systems: An Overview of Issues, Methodologies, and FSAP Experiences. International Monetary Fund (IMF) Working Papers, January 1988.

Basel Committee on Banking Supervision (2003). Overview of the New Basel Accord. Bank for International Settlements (BIS), July 2003.

Borio, C., C. Furfine, and P. Lowe, (2001). Procyclicality of the Financial Systems and Financial Stability. BIS Papers No. 1, March 2001.

Boss, M. (2002). A Macroeconomic Credit Risk Model for Stress Testing the Austrian Credit Portfolio. Financial Stability Report 4, Oesterreichische National Bank, October 2002.

Crouhy, M, D Galai and R Mark (2000). A Comparative Analysis of Current Credit Risk Models. Journal of Banking & Finance, Vol. 24, No. 1, January 2000.

Hoggarth, G., S. Sorensen, and L. Zilcchino. (2005). Stress Tests of UK Banks Using a VAR Approach. Bank of England Working Paper No. 282. November 2005.

International Monetary Fund (IMF). Canada: Financial System Stability Assessment Update. IMF Country Report No. 08/59, February 2008.

Kalirai, H. and M. Scheicher (2002). Macroeconomic Stress Testing: Preliminary Evidence for Austria. Austrian National Bank Financial Stability Report No. 3, May 2002.

Misina, M and D. Tessier (2007). Sectoral Default Rates Under Stress: Importance of Non- Linearity. Bank of Canada Financial System Review, June 2007.

Misina, M., D. Tessier and S. Dey (2006). Stress Testing the Corporate Loan Portfolio of the Canadian Banking Sector. Bank of Canada Working Paper, No. 47, 2006.

Navarrete, E. (2006). Practical Calculation of Expected and Unexpected Losses in Operational Risk by Simulation Methods. MPRA Paper 1369, University Library of Munich, Germany. 2006.

Pesola, J. (2001) The Role of Macroeconomic Shocks in Banking Crises. Bank of Finland Discussion Paper No. 6, April 2001.

Servigny, A. and O. Renault (2004). Measuring and Managing Credit Risk. New York: McGraw-Hill Companies, 2004.

Sorge, M. and K. Virolainen (2006). A Comparative Analysis of Macro Stress-Testing Methodologies With Application to Finland, Journal of Financial Stability Vol. 2, No. 2, June 2006.

Valentinyi- Endrész, M. and Z. Vásáry (2008) Macro stress testing with sector specific bankruptcy models. Hungarian National Bank Working Papers, No. 2, 2008.

Virolainen, K. (2004). Macro Stress Testing with a Macroeconomic Credit Risk Model for Finland. Bank of Finland Discussion Paper No. 18, October 2004.

Vlieghe, G. (2001). Indicators of fragility in the UK corporate sector. Bank of England Working Paper No. 146, December 2001.

Wilson, T.C. (1997). Portfolio Credit Risk (I). Risk, September 1997. (Reprinted in Credit Risk Models and Management. 2004, 2nd edition, edited by David Shimko, Risk Books).

• Business Bankruptcies for 1997-Q1 to 2007-Q4.
  -Source: Office of the Superintendent of Bankruptcies (OSB) via Statistics Canada (Table 177- 0004).
  -Available monthly, by province, by industry (NAICS15)

• Establishment Counts for 1997-Q1 to 2007-Q4
  -Source: Statistics Canada- Canadian Business Patterns
  -Available semi-annually, by province, by industry (NAICS)

• Default Rate for 1997-Q1 to 2007-Q4 
  -Convert monthly bankruptcy data to quarterly by adding the three months that constitute a quarter.
  -Convert semi-annual establishment count data to quarterly by averaging the two adjacent semi-annual values to find the missing quarters value.
  -Divide quarterly bankruptcy figures by corresponding establishment counts

• Macroeconomic Variable for 1997-Q1 to 2007-Q4
  o Gross Domestic Product (GDP) for Canada
  -Source: Statistics Canada (Table 379-0027)
  -Available monthly at basic prices, by industry (NAICS), in chained 2002 dollars.
  
  o Real Interest Rates (R-Can and R-US) for Canada and USA Canada- Chartered
  bank administered prime business rates
  -Source: Bank of Canada via Statistics Canada (Table 176-0043)
  -Available monthly as a percent, converted into real dollars USA- US bank
  prime loan business
  -Source: Bank of Canada via Statistics Canada (Table 176-0044)
  -Available monthly as a percent, converted into real dollars.
  
  o Commodity Price Index
  -In US dollar terms, indexed so 1982-1990 equal 100
  -Source: Statistics Canada (Table 176- 0001)
  -Available monthly

• Sectoral Exposure for 1997-Q1 to 2007-Q4
  -Chartered bank classification of non-mortgage loans (domestic currency)
  -Source: Office of the Superintendent of Financial Institutions via Statistics Canada (Table 176-0016)
  -Available quarterly, by industry (NAICS) in Canadian dollars.

• Regional Exposure for 1997-Q1 to 2007-Q4
  -Chartered banks regional distribution of assets and liabilities- Total business loans
  - Source: Bank of Canada via Statistics Canada (Table 176-0045)
  -Available quarterly, by province in Canadian dollars

• Loan Loss Provision 1997-Q1 to 2007-Q4
  -Individual allowance for impaired non-mortgage loans divided by total non-mortgage loans made to individuals for business purposes (converted into a percent)
  -Available for Bank of Montreal, Royal Bank of Canada and Toronto Dominion Bank
  -Source: Office of the Superintendent of Financial Institutions (OSIF)- the Consolidated Balance sheet and Impaired Assets statement for each individual bank.

15 NAICS‐ North American Industry Classification System.