Causal Insight into CAPM Class Models (Part III)

Interpreting the CAPM Equation

Social scientists often interpret regression results in terms of causal processes, even if the mathematics of regression is really just about quantifying relationships observed in data. Since active managers often look for models to predict returns, I initially thought of CAPM as mechanism to predict returns, as the name ”expected return“ implies. In causal modeling terms, CAPM is simply a regression model where a dependent variable (expected asset returns) is assumed to be predicted by an independent variable (market or systemic returns).

So, if one were to ask, ”what caused the return on asset A to be Y%?“ the answer under CAPM would be:

”The Market as a whole returned X% above the Rfr, and that caused the Asset to return (Beta*X%) more than the Rfr. The difference, [Y% - (Beta*X%)-Rfr] was caused by other [presumably unpredictable] idiosyncratic factors.“

This is all true, provided that you accept that the relationship between A and the Market is a causal one. This is fairly defensible on logical grounds, since if Beta is statistically significant, there is most likely a relationship between the variables and that relationship is much more likely to run from the Market to the asset than the other way around (it would be a magical Asset to drive the rest of the Market). Technically, it is possible that some third factor could drive both asset A and the Market, but if we don’t actually know what that other factor is, the Market is probably just as good a proxy for that factor as anything else. So the causal interpretation that the Market made the Asset return (Beta*X%) is plausible on first brush.

This logic works, but there is a big practical problem. If we are using the return on the Market to predict the return of Asset A, the problem is that we still do not know what the Market will actually return in the same time period. From an asset manager’s forecasting viewpoint, both the Asset return and the Market return are in the future and essentially unpredictable, other than (perhaps) their long term averages. Essentially CAPM is asserting that one thing that we don’t know yet (Asset A’s return in the future) is consistently caused by something else that we don’t know yet (the Market’s return in the future). It may be true, technically, but it doesn’t really help us if we don’t have a way to predict what the Market is going to return.

While there are no solutions to the problem of not knowing the future, there are some ways to approach it:

1) If one i) has a long time horizon, and ii) is interested in long term averages, and iii) believes that the historical past is reasonably representative of the long term future, it is possible to use CAPM-type models to estimate long-term average returns for an asset. For many purposes, this is simple and sufficient.

2) If one can develop a separate model that predicts overall market returns reasonably well - such as a time series regression or ARIMA - one can use that model as an input into the asset return model. This would increase the apparent overall risk of the asset pricing model, since the uncertainties of the asset model’s output uncertainty would be compounded by the the fact that one of the inputs (expected market return) is also uncertain. However, the total risk of using both models together is likely to be lower, because the market expectations model - if useful - should reduce the uncertainty in market expectations over the uncertainty of not having any expectations at all. With this approach, there is also model risk from two separate models: the asset pricing model and the market return model.

3) If one has an asset model and a market expectations model as in approach (2), it might be more efficient to try to combine the independent variables into one single model to predict returns on the asset. This would improve predicted asset returns if there is any correlation between an asset’s beta parameter and any of the factors that determine market expected returns.

4) One can try to eliminate the effect of market uncertainty by taking appropriately balanced and opposite positions in the asset and the market as a whole. For example, for every X% of long exposure in an asset, one can take Beta*X% short exposure in the market index, through ETFs, index futures, or in some cases the index itself. Provided that the returns model is a good one and is still appropriate, this has the effect of removing the uncertainty that comes with the market performance, leaving an investor with only the non-market or non-systematic component of an asset’s risk.

Causality, CAPM, and beyond

Approach (4) has some interesting implications for returns modeling. If one believes wholeheartedly in CAPM, then taking opposing positions in an asset and the market index such that net Beta=0 will effectively remove all systemic risk from a position and leave the investor with asset-specific (a.k.a diversifiable or idiosyncratic) risk only. In a pure CAPM world, there isn’t much advantage to this, because the expected value of asset-specific risk is zero (as discussed in Part 1), and the volatility created by assuming this risk will actually eat away at a portfolio over time.

The above assumes that the idiosyncratic component of asset risk is completely unpredictable, however. In Bob Litterman’s recent discussion of ”exotic beta“, he suggests that the returns to non-systemic factors such as size or value might average out to zero over the long term, yet still offer predictive value in the short term. Similarly, fundamental or competitive analysis might identify one company as riskier or less risky depending on something like competitive position, cash reserves, a valuation figure, or some other temporary and observable factor. In these cases, although the expected return on asset-specific risk is zero over the long term, it can be related to observable predictors in the short or medium term.

Believing that there are observable predictors for short term idiosyncratic returns requires believing that markets are not fully efficient - that it takes time for observable predictors to be disseminated and incorporated into market prices, and therefore presently known factors can predict some level of future returns. Moreover, there are empirical reasons to believe that markets are not fully efficient, such as the presence of momentum / autocorrelation in some markets.

If we assume that markets are not fully efficient, and that some observable variables can help in predicting short-term returns, then the use of a market factor in a returns model has another key role. Under these conditions, we have a model that states that some parts of an asset’s return are predictable and some parts are not. The effect of the predictable component may be small compared to the effect of the unpredictable component, but we would still like a way to incorporate this into an investment decision. If we have a variable, X, that we think can explain some part of a return, the model now looks like:

R_a = Rfr + (Beta_m) * (R_m) + (Beta_X) * X(t-1) + noise

Where R_a is the return on the asset, R_m is the return on the market, Rfr is the risk-free rate, and X(t-1) is the variable we think can predict returns. Variable X is measured at time=(t-1) because we know its value at the start of the investment period, but we can only receive the returns from an asset or the market at the end of this period.

The equation form can obscure the causal structure of the model. A causal diagram helps to make the model clearer.


[If not embedded, click here for causal diagram]

The model says that the asset return R_a is the result of a knowable variable X, and an unpredictable market return R_m. This means that the model says that variable X has a (Beta_X) effect on the asset’s return after controlling for the impact of (more or less) unpredictable market moves.

Isolating the value of investment insights

Including the (unpredictable) market return into the causal diagram has two implications, one statistical, the other financial. Statistically speaking, if variable X is uncorrelated with market returns, it is not strictly necessary to include the market return in the model, because a regression process should be able to pick up the statistical significance of X and estimate it’s beta in any case. However, the market return is still useful as a method of reducing statistical uncertainty and getting a more accurate estimate of X’s effect. The technical term for this is that the market variable functions as an instrumental (rather than causal) variable: we may not be able to predict the market return, but in the regression it reduces the quantity of unexplained variance.

In addition, it is statistically helpful to include the market return in the predictive model because there is still some chance that X is mildly correlated with the market return and so including the market return in the same period will control for that possibility.

Financially speaking, the advantage of this formulation is that the effects of the market return can be hedged so that a portfolio produces only the returns achieved from knowing X and not returns from assuming market risk. This is useful because the returns to variable X may be quite small compared to the overall effect of the entire market and might otherwise be difficult to capture.

For example, if variable X suggests that asset A is relatively attractive, that may be true, comparatively speaking, but even if the asset is better than its alternatives, the market as a whole may still go down, taking asset A with it and destroying any benefit of your insight into X. However, by shorting the correct number of market futures or shares of an index ETF, the effects of market uncertainty can be eliminated from the portfolio, and the effects of variable X isolated. This can be an attractive way of ensuring that your investment positions capture your insight into variable X independent of what the market as a whole does and maximizes the value of your insight. It does, of course, depend on your model of returns being both accurate and stable. In times of great market turmoil -- like the present -- models based on past performance data become less reliable and many fail completely.

Recap

What I hoped to show in this blog piece is how the specification of returns models connects to causal analysis, and why -- even if we cannot predict what the market return will be -- a regression-based returns model should still include a contemporaneous market factor. The main reasons are:

1) It produces more efficient (i.e. precise) estimates of the effects of other variables that you can (potentially) use for predictive modeling.

2) It controls for the possibility that your predictive variables might be somewhat correlated with future market wide returns.

3) The beta estimate for the market return factor can be used to hedge out the effects of the market return on your investment, leaving you with a position that capitalizes on the quality of your insight. Since the market return is often the dominant factor in the performance of individual assets, this can be very useful for managing the total risk one takes by acting on a specific insight (in this case, the predictive value of variable X).

Causal Insight into CAPM Class Models (Part III)

Interpreting the CAPM Equation

Social scientists often interpret regression results in terms of causal processes, even if the mathematics of regression is really just about quantifying relationships observed in data. Since active managers often look for models to predict returns, I initially thought of CAPM as mechanism to predict returns, as the name ”expected return“ implies. In causal modeling terms, CAPM is simply a regression model where a dependent variable (expected asset returns) is assumed to be predicted by an independent variable (market or systemic returns).

So, if one were to ask, ”what caused the return on asset A to be Y%?“ the answer under CAPM would be:

”The Market as a whole returned X% above the Rfr, and that caused the Asset to return (Beta*X%) more than the Rfr. The difference, [Y% - (Beta*X%)-Rfr] was caused by other [presumably unpredictable] idiosyncratic factors.“

This is all true, provided that you accept that the relationship between A and the Market is a causal one. This is fairly defensible on logical grounds, since if Beta is statistically significant, there is most likely a relationship between the variables and that relationship is much more likely to run from the Market to the asset than the other way around (it would be a magical Asset to drive the rest of the Market). Technically, it is possible that some third factor could drive both asset A and the Market, but if we don’t actually know what that other factor is, the Market is probably just as good a proxy for that factor as anything else. So the causal interpretation that the Market made the Asset return (Beta*X%) is plausible on first brush.

This logic works, but there is a big practical problem. If we are using the return on the Market to predict the return of Asset A, the problem is that we still do not know what the Market will actually return in the same time period. From an asset manager’s forecasting viewpoint, both the Asset return and the Market return are in the future and essentially unpredictable, other than (perhaps) their long term averages. Essentially CAPM is asserting that one thing that we don’t know yet (Asset A’s return in the future) is consistently caused by something else that we don’t know yet (the Market’s return in the future). It may be true, technically, but it doesn’t really help us if we don’t have a way to predict what the Market is going to return.

While there are no solutions to the problem of not knowing the future, there are some ways to approach it:

1) If one i) has a long time horizon, and ii) is interested in long term averages, and iii) believes that the historical past is reasonably representative of the long term future, it is possible to use CAPM-type models to estimate long-term average returns for an asset. For many purposes, this is simple and sufficient.

2) If one can develop a separate model that predicts overall market returns reasonably well - such as a time series regression or ARIMA - one can use that model as an input into the asset return model. This would increase the apparent overall risk of the asset pricing model, since the uncertainties of the asset model’s output uncertainty would be compounded by the the fact that one of the inputs (expected market return) is also uncertain. However, the total risk of using both models together is likely to be lower, because the market expectations model - if useful - should reduce the uncertainty in market expectations over the uncertainty of not having any expectations at all. With this approach, there is also model risk from two separate models: the asset pricing model and the market return model.

3) If one has an asset model and a market expectations model as in approach (2), it might be more efficient to try to combine the independent variables into one single model to predict returns on the asset. This would improve predicted asset returns if there is any correlation between an asset’s beta parameter and any of the factors that determine market expected returns.

4) One can try to eliminate the effect of market uncertainty by taking appropriately balanced and opposite positions in the asset and the market as a whole. For example, for every X% of long exposure in an asset, one can take Beta*X% short exposure in the market index, through ETFs, index futures, or in some cases the index itself. Provided that the returns model is a good one and is still appropriate, this has the effect of removing the uncertainty that comes with the market performance, leaving an investor with only the non-market or non-systematic component of an asset’s risk.

Causality, CAPM, and beyond

Approach (4) has some interesting implications for returns modeling. If one believes wholeheartedly in CAPM, then taking opposing positions in an asset and the market index such that net Beta=0 will effectively remove all systemic risk from a position and leave the investor with asset-specific (a.k.a diversifiable or idiosyncratic) risk only. In a pure CAPM world, there isn’t much advantage to this, because the expected value of asset-specific risk is zero (as discussed in Part 1), and the volatility created by assuming this risk will actually eat away at a portfolio over time.

The above assumes that the idiosyncratic component of asset risk is completely unpredictable, however. In Bob Litterman’s recent discussion of ”exotic beta“, he suggests that the returns to non-systemic factors such as size or value might average out to zero over the long term, yet still offer predictive value in the short term. Similarly, fundamental or competitive analysis might identify one company as riskier or less risky depending on something like competitive position, cash reserves, a valuation figure, or some other temporary and observable factor. In these cases, although the expected return on asset-specific risk is zero over the long term, it can be related to observable predictors in the short or medium term.

Believing that there are observable predictors for short term idiosyncratic returns requires believing that markets are not fully efficient - that it takes time for observable predictors to be disseminated and incorporated into market prices, and therefore presently known factors can predict some level of future returns. Moreover, there are empirical reasons to believe that markets are not fully efficient, such as the presence of momentum / autocorrelation in some markets.

If we assume that markets are not fully efficient, and that some observable variables can help in predicting short-term returns, then the use of a market factor in a returns model has another key role. Under these conditions, we have a model that states that some parts of an asset’s return are predictable and some parts are not. The effect of the predictable component may be small compared to the effect of the unpredictable component, but we would still like a way to incorporate this into an investment decision. If we have a variable, X, that we think can explain some part of a return, the model now looks like:

R_a = Rfr + (Beta_m) * (R_m) + (Beta_X) * X(t-1) + noise

Where R_a is the return on the asset, R_m is the return on the market, Rfr is the risk-free rate, and X(t-1) is the variable we think can predict returns. Variable X is measured at time=(t-1) because we know its value at the start of the investment period, but we can only receive the returns from an asset or the market at the end of this period.

The equation form can obscure the causal structure of the model. A causal diagram helps to make the model clearer.


[If not embedded, click here for causal diagram]

The model says that the asset return R_a is the result of a knowable variable X, and an unpredictable market return R_m. This means that the model says that variable X has a (Beta_X) effect on the asset’s return after controlling for the impact of (more or less) unpredictable market moves.

Isolating the value of investment insights

Including the (unpredictable) market return into the causal diagram has two implications, one statistical, the other financial. Statistically speaking, if variable X is uncorrelated with market returns, it is not strictly necessary to include the market return in the model, because a regression process should be able to pick up the statistical significance of X and estimate it’s beta in any case. However, the market return is still useful as a method of reducing statistical uncertainty and getting a more accurate estimate of X’s effect. The technical term for this is that the market variable functions as an instrumental (rather than causal) variable: we may not be able to predict the market return, but in the regression it reduces the quantity of unexplained variance.

In addition, it is statistically helpful to include the market return in the predictive model because there is still some chance that X is mildly correlated with the market return and so including the market return in the same period will control for that possibility.

Financially speaking, the advantage of this formulation is that the effects of the market return can be hedged so that a portfolio produces only the returns achieved from knowing X and not returns from assuming market risk. This is useful because the returns to variable X may be quite small compared to the overall effect of the entire market and might otherwise be difficult to capture.

For example, if variable X suggests that asset A is relatively attractive, that may be true, comparatively speaking, but even if the asset is better than its alternatives, the market as a whole may still go down, taking asset A with it and destroying any benefit of your insight into X. However, by shorting the correct number of market futures or shares of an index ETF, the effects of market uncertainty can be eliminated from the portfolio, and the effects of variable X isolated. This can be an attractive way of ensuring that your investment positions capture your insight into variable X independent of what the market as a whole does and maximizes the value of your insight. It does, of course, depend on your model of returns being both accurate and stable. In times of great market turmoil -- like the present -- models based on past performance data become less reliable and many fail completely.

Recap

What I hoped to show in this blog piece is how the specification of returns models connects to causal analysis, and why -- even if we cannot predict what the market return will be -- a regression-based returns model should still include a contemporaneous market factor. The main reasons are:

1) It produces more efficient (i.e. precise) estimates of the effects of other variables that you can (potentially) use for predictive modeling.

2) It controls for the possibility that your predictive variables might be somewhat correlated with future market wide returns.

3) The beta estimate for the market return factor can be used to hedge out the effects of the market return on your investment, leaving you with a position that capitalizes on the quality of your insight. Since the market return is often the dominant factor in the performance of individual assets, this can be very useful for managing the total risk one takes by acting on a specific insight (in this case, the predictive value of variable X).

Whence the Market Hangover

It’s been about two weeks since financial armageddon started - it really feels much longer than that - and I’ve been feeling that I’m way overdue for some commentary on political and financial events. What a great opportunity to combine my political analysis training with my more recent emphasis on financial practice.

Of course, the market situation changes so quickly that one of the biggest challenges is that the market can look completely different between the start and the end of even a single blog piece, which can make it difficult to figure out what to write that can add value when readers stop by.

In these moments, I find that it helps to step back to get a look at the big picture. In the middle of the firestorm, it’s hard to predict which axe will fall next, but we can try to get a sense of how we got here and whether turning, pushing forward, or retracing our collective steps is most practical.

The big issue is that we appear to be at the bursting of what some called “a bubble in risky assets,” which one might call “the omnibubble.“ Risk assets or risky assets are simply things that can gain or lose value - stocks, bonds with default risk, commodities, real estate, etc. - pretty much any investible asset other than cash or short-term treasury securities. Unlike the tech bubble or the housing bubble, the risky asset bubble is harder to see, because it happens to virtually all investible assets simultaneously - hence the term ”omnibubble.“ From a returns perspective, there is no relative “horizon” one can use to orient oneself. The only real gauge one might have used is to compare recent returns on all assets with historical returns, but so many things have changed in the economy over the last 20-25 years that it was easy for people to say the magic words, “this time it’s different.”

In this case, the risky asset bubble was made possible by a “credit bubble” in which the cost of borrowing was extremely cheap by historical standards, and also inexpensive compared to the level of risk that the lenders took by lending. What this means practically is that with easy borrowing terms, individual and institutional investors bought more risk assets than they would otherwise, and this extra demand made the price of virtually all assets rise tremendously - stocks, bonds, real estate, commodities, etc.. It was essentially a bubble in all asset classes simultaneously. In addition, easy credit terms allowed consumers and businesses to purchase more, so the higher asset prices may have seemed to be justified by the growth in revenues from credit-supercharged consumers and businesses.

The seduction of easy credit and illiquid assets

When you are leveraged (i.e. buying suff without having to pay the full price up front), rising asset prices create magnified returns. This means that individuals and investors in an upswing may feel richer and safer than one might think by looking at the raw asset price increases alone. For example, a homeowner who saw their home price increase by 10% would have their home equity appear to increase by more than 50% if they have an 80% loan-to-value mortgage. This number would appear very large because many homeowners likely looked only their home equity values at a single point in time rather than subtracting their interest payments from equity gains to have a picture of total returns. Even assuming a mortgage rate of 5% on a loan, a home owner’s total return would still be on the order of 30%, and the home equity values would also look higher because of principal repayments (unless the loan was interest only). With some loans approaching 100% loan-to-value, interest only, and low-teaser rate adjustable mortgages, the return on equity investments would start to look astronomical. Add in the fact that these semi-illusory returns of 30-50% were on homes worth several times the average working salary, and a wide variety of people were feeling nicely wealthy, with plenty available to consume, even as the diversity of employment opportunities in the general economy were starting to diminish.

The wealth that this leverage created made people feel richer, smarter, and, more importantly, safer than they really are. With such outstanding paper profits, many felt they could support higher consumption and riskier investments than they really could, because 1) the equity returns were magnified by tremendous leverage from loans, 2) people underestimated how quickly that cushion could disappear in a downturn, and 3) housing prices are illiquid because it takes time to prepare and sell a house, including the due diligence required, so housing price changes appear to rise continuously and deceptively smoothly.

Even more, homeowners were seeing 30-50% increases in their home equity even though home prices were rising only 10%. That kind of differential may well have left consumers feel not only that they were well-off, but also that they were smarter than the crowd. After all, they had taken a loan like everyone else, and yet their perceived wealth had increased several times more than the average home price increase.

All of this lulled people into consuming and borrowing more than is really sustainable. The returns from leveraged assets were extremely concrete, and much of the risks were abstract, if not hidden. Even relatively careful people might start cautiously, find that their “investment” worked, increase their bet next time, and start to figure that they understood the system. The outrageous “success” of less cautious neighbors might have made smart, cautious people feel like fools, seducing them into taking larger risks to keep up. Feeling safer and more in control, people consumed more discretionary items, propping up business revenues and contributing to the illusion that the high asset prices were justified by growth rates. To diversify, some housing profits went into other risky assets, pushing up those asset prices.

The danger of leverage is that the magnification of upside returns also happens on the downside. Just as a 10% rise in home prices - if leveraged - may lead to 30-50% increases in home equity wealth, a 10% drop will produce 30-50% collapses. But ordinary homeowners and house-flippers may not have intuitively absorbed this. In a rising market, homeowners may check their equity regularly and be happy with their 30%+ results. In a decline, they may stop looking, and hear that housing prices have dropped a few percent. Unless they are trained professionals, their risk assessment is likely that they have upsides of 30-50%, downsides of around 10%, and vastly underestimate the risk they take.

From whence all this credit?

All that easy credit in turn came from new financial technologies which also appeared to make things safer than they really are. Lenders traditionally need to charge interest rates that compensate the lender for the risk that the borrower might default. The technique of “asset securitization” lowered the cost of credit (i.e. interest rates for mortgages, credit cards, and many other things) four ways:

1) By “chopping up” mortgages or other loans into small enough pieces, securitization allowed a larger number of investors essentially to become mortgage lenders, because one could purchase a moderately priced security instead of requiring several hundred thousand dollars to make a single loan. That meant a lot more people could purchase mortgages, and the demand (driven by investor desires to diversify asset exposure) meant that one could find enough buyers for mortgages without raising rates as high would be necessary with fewer available investors .

2) Along with chopping up mortgages into securities, “tranching” allowed firms to shift around who theoretically bears the risk of mortgage defaults. This process can theoretically match mortgage securities better to the needs of different investors, create a larger number of buyers through “customization,” and thus reduce the interest rates on loans.

3) By “diversifying mortgage pools” and bundling lots of of mortgages together, a mortgage security should theoretically reduce the degree an investor is exposed to any one loan. As long as the chance of one borrower defaulting is unrelated to the chance of another defaulting, this diversification reduces the risk of investing in a mortgage and should therefore lower the required interest rates for mortgages.

4) By “enhancing” the securities with “credit default swaps” or CDSs - a kind of insurance for bad payments - these securitized payments can look more secure than they would otherwise be. Since the CDS seller will step in with money if one of the underlying mortgages defaults, this lowers the required rate for investors to give borrowers their money.

Securitization and their enhancement with CDSs has a lot to do with the inexpensively available credit that puffed up the risky asset bubble. Some of securitization’s benefits are genuine, and so in the future, it’s important not to throw the baby out with the bathwater and abandon securitization forever. The main problems with securitization is that 1) analysts didn’t agree on how to value securitized assets, 2) it was often very difficult to know what assets (specific morgtages, borrower’s credit quality etc) were actually bundled together in a particular security, and 3) the CDSs had more counterparty risk (the risk that the insurer would not pay up when needed) than shown in the models.

Simply put, this meant that securitizers did not have a good view of the risk they were taking and therefore were driven by competition to come up with interest rates that were too low to compensate for the level of risk actually undertaken. The availability of these low interest rates fed into the expansion of credit more generally and the leveraging that created the risky asset bubble.

Mortgage Securitization and Perverse Incentives for Mortgage Originators

One knew that the housing bubble was inflating rapidly when we saw no-documentation, no-money-down, mortgages being offered. Clearly lending money to people with no evidence that they could pay and holding on to the hope that asset prices will rise sufficiently to cover any losses was a fools game for the lenders. But how could this happen, and why?

Under normal lending conditions, the lender has an incentive to perform due diligence on the borrower - check if there is income to make payments, current debt levels, asset reserves, equity reserves etc.. As mortgage securitization expanded, it created a layer of obscurity between the ones underwriting a mortgage (buyers of the securities) and those originating the mortgages (the mortgage officer that borrowers talked to). The mortgage officer would make a loan, collect an origination fee on the loan, and then almost immediately sell the cash flows (and risk of their default) to a securitization firm (a.k.a. special purpose vehicle), who would bundle the mortgage into a large pool with hundreds or thousands of other mortgages to achieve diversification.

In short, the mortgage brokers who were telling people that they could afford adjustable mortgages with low monthly payments and no documentation of income, assets, etc. had almost *NO* incentive to find good borrowers, and every incentive to originate more loans. As more and more high-quality borrowers were installed in homes with mortgages, mortgage brokers turned to lower and lower quality borrowers, bundling them together in securities and enhancing their credit through insurance CDS.

Those who buy the securities and are on the hook for default risk still have an incentive to do due diligence on the underlying mortgages, but now the process is much much more difficult (and obscured). Instead of doing due diligence on a single borrower, using historical norms and comparisons, the lender now needs to do due diligence on all the borrowers in the security. If part of the point of the security was to be able to get paid for taking on mortgage-type risks without having to fund the whole 100% of the mortgages, then the cost of doing that due diligence on all the underlying borrowers is now enormous compared to the size of the investment. It was essentially impractical for mortgage-backed-securities buyers to do due diligence. Instead, lenders trusted ratings agencies to perform the due diligence and rate the securities.

Now ratings agencies have a conflict of interest: they are paid by the organizations they rate, yet must be objective with their ratings. With most corporations, the fact that they almost always need a rating to attract lenders is enough to keep the conflict of interest in check. If GM can’t get a good rating from one of the agencies, there are not many alternatives for GM other than to float unrated debt - which will be expensive. But special purpose vehicles (SPVs) can be created and dissolved quickly, and their assets reconstituted. If a ratings agency does not give good ratings to an SPV, it can be dissolved, reconstituted, and take its business elsewhere. Therefore the conflict of interest is more severe.

Add to this that many of these securities had mortgages of new kinds of borrowers for which there is little historical precedent and data. Historically, lenders had not been giving mortgages to people with no income information, assets, and 100% loan-to-value ratios, and so there was little historical data on which to gauge the riskiness. This meant that ratings agencies had to stress-test their financial models with many numbers simply picked up out of the air. The securities had ratings of their creditworthiness, but nobody knew whether the models to give them these ratings were any good. In this environment, the temptation to pick slightly rosier numbers out of the air to retain clients and rate the next SPV seemed both harmless and economically urgent.

Like the schoolyard game of “telephone,” information about the underlying mortgages would tend to get garbled and fuzzy as it moved from borrower, to orginator, to SPV, to ratings agency, to lender. Not only that, but the garbling was biased: bad news was very likely deliberately garbled to help pass the securities on to the next borrower.

Starting with Mortgages, moving to other assets

This discussion centers on how the securitization of mortgages artificially lowered borrowing rates by creating a long string of intermediaries that had the effect of obscuring and ultimately understating the risk of these assets. With the risk obscured, competitive pressures could lower the interest rates required to finance homes. This financial technique was then applied to auto loans, credit card loans, and virtually any kind of debt obligation one could construct.

The artificially low rates in turn allowed many people to buy even more assets on borrowed money - stocks, real estate, commodities, consumer goods - and drove their price up. The price running upwards seemed to confirm the wisdom of borrowing, and borrowing itself magnified the gains in equity. This process began over a decade ago and accelerated greatly in recent years.

Now we feel the downside. At this point lenders are tapped out, assets need to come down in price to sell, and the pain of declining levered assets is at least as bad as the euphoria of levered gains. The drop in asset prices is levered to deal crushing blows to their holders, who are often forced to sell. Those sales depress asset prices further, forcing others to sell. Pretty soon large numbers of people need to sell assets to raise cash to cover their loans, but everyone is selling more or less the same stuff at the same time, creating a glut that crushes prices.

Now people start to get frightened. Even those who did not lever start to see asset prices declining rapidly and want to sell. Banks are threatened with runs on their reserves. Nobody is able to borrow, so they try to sell to raise cash. Nobody wants to lend, because the risk of not being paid back is suddenly very high.

Conclusion, but not the end.

Summing up this long piece, what we see is that we have a bursting omnibubble in virtually all risky assets. These assets were inflated by extremely inexpensive credit that allowed investors to buy with borrowed money and consumers to consume with borrowed money. The money to borrow was made available through securitization practices that obscured (often deliberately) the degree of risk to the lender. The person who actually decided to make the loan had no real incentive to do due diligence on the borrower, making the money available on exceedingly favorable terms that are not enough to compensate lenders for the true risks involved. Thus, when mortgages started defaulting (and this is likely just the beginning of a cycle of defaults that will include auto loans and credit cards), these securities started looking substantially less valuable than they had been, even if the rate of mortgage default was low. Moreover, companies who provided CDS insurance as a “credit enhancement” did not have sufficient cash reserves to make insurance payments, further pushing down the price.

After several decades of borrowing on inexpensive terms, it looks like we are going to have to collectively make large principal and interest payments for all the excess consumption we have gotten so inexpensively (up to now). There may be some choice as to whether we pay this in higher taxes (pay the government to take care of it), higher inflation (we earn the same but buy less with it), or a depression (simply halt the economy for a while). Hopefully we won’t have all three.