SeekerBlog

TIE: So how should economists operate in a world of imperfect knowledge?

Frydman: Economists are trained early on to believe that models that do not generate exact predictions are not worthy of consideration. But the opposite is true. To be useful, economic models must be consistent with the basic fact that participants hold diverse views about the future. How this diversity translates into prices over time must be left for the markets to determine. No mathematical model can hope to mimic exactly what markets do.

A worthwhile interview in International Economy with Roman Frydman, related to the new book Imperfect Knowledge Economics: Exchange Rates and Risk.

Frydman & Goldberg offer policy proposals that merit careful consideration. One is to enhance the price discovery for assets rated by S&P, Moody’s etc. Today the ratings agencies publish one value for each entity — hidden behind that rating are stress tests of unknown quality.

What if the agencies published at least two values for each company? E.g., the current central tendency value, plus say a “minus 3 standard deviation” rating? Frydman is proposing a more nuanced suite of ratings on each company — varying +/- about the expectation.

What I find intriguing here is that Frydman may be suggesting a more efficient way for pricing complex assets such as CDOs — simply by mandating more information be provided to market players by the rating agencies.

I.e., instead of top-down regulation, let us try information transparency ==>> feeding data to markets which do the price discovery. Brilliant, I think:

…TIE: Your book leads an assault against the rational expectations school of economics. What is wrong with rational expectations?

Frydman: To put it simply, despite its name, Rational Expectations Hypothesis (REH) has nothing to do with rational forecasting.

The basic problem with REH models can be traced back to the widespread belief among contemporary economists that economic models can’t be considered scientific unless, except for a purely random error term, they generate exact predictions of outcomes. Obviously, forecasts of asset prices drive outcomes in financial markets. But if we acknowledge this, and yet insist on exact predictions, we must come up with an exact model of how market participants think about the future. That’s a tall order. And REH goes even further. It supposes that market participants forecast according to the exact model that economist themselves create.

This is even odder than it sounds. After all, economists’ bread and butter is the search for alternative models. But if a particular economist’s model were to capture rational forecasting, the use of other models would be irrational,

Interestingly, John Muth, who proposed REH in the early 1960s, was well aware of the danger that the term “rational expectations” might suggest some notion of individual rationality. But, despite his warning, REH is commonly viewed as the way to model rational forecasting. As such, REH is not just of academic interest. It is widely used as the centerpiece of sophisticated models to price derivative products and risk in financial markets.

TIE: So is not the failure of sophisticated finance models the telltale that RICH is seriously flawed?

Frydman: Spot on. We should have expected REH models to fail in financial markets. After all, outcomes in these markets are primarily driven by forecasts, and REH is a particularly poor forecasting model. Many of our colleagues cite REH models’ failures as evidence that markets are populated by irrational traders. In fact, rational individuals in real-world markets do not follow pre-existing rules and procedures, let alone exact forecasting rules implied by economists’ models. So the failures of REH models do not prove that markets are irrational. They simply show that economists have a wrong model of rationality. Neither REH models nor, for that matter, any other quantitative model, can capture exactly how profit-seeking market participants forecast the future.

TIE: You argue that economic forecasting is not a science. The forecaster is more like an entrepreneur who relies on intuition and judgment. Where do quantitative methods fit in?

Frydman: Quantitative models have a place in the forecaster’s toolbox, but the choice among them and how they are used must involve subjective judgments. To begin with, quantitative models contain a number of unknowns that must be estimated. But, because an economy’s structure changes, it is unclear how much past data a forecaster should use to come up with statistical estimates of the unknowns in his model. Even the most sophisticated statistical techniques would not automatically pinpoint when the last structural break had occurred. Of course, the choice among various alternative models also requires subjective judgments. Unsurprisingly, even when it comes to the past, let alone forecasting the future, interpretations vary among individuals depending on their personal knowledge, experience, and intuitions.

Consider the problem of forecasting exchange rates. Many market participants no doubt use quantitative models to form exact forecasts of the future exchange rate, for example, that in a week a euro will cost $1.50. After all, currency traders must decide on their market position at each point in time. But, although they may base their trading on exact predictions, they do not arrive at such predictions by relying solely on quantitative models, much less the same model in every time period. Instead, they often combine quantitative models with their own insights concerning other traders’ behavior, the historical record on exchange rate fluctuations, and their evaluation of the impact of past and future decisions by policy officials. And, because they act on the basis of different experiences, interpretations of the past, and intuitions about the future, they adopt different strategies for forecasting exchange rates.

TIE: So how should economists operate in a world of imperfect knowledge?

Frydman: Economists are trained early on to believe that models that do not generate exact predictions are not worthy of consideration. But the opposite is true. To be useful, economic models must be consistent with the basic fact that participants hold diverse views about the future. How this diversity translates into prices over time must be left for the markets to determine. No mathematical model can hope to mimic exactly what markets do.

For example, euro bulls and bears have diametrically opposed forecasts. Despite drastic differences in how they form their forecasts, however, the ways in which they revise their forecasting strategies may share certain qualitative features. Imperfect Knowledge Economics (IKE) makes use of such regularities and shows how they help us understand the tendency for the exchange rate to move away from parity in some periods of time and revert back to this benchmark in others. Such qualitative predictions are admittedly less ambitious than aiming for exact predictions. But our book shows that economists can learn more about markets if we ask less of our models.

TIE: How would you apply your approach to understanding currency movements?

Frydman: IKE acknowledges the limits of economists’ knowledge. Not aiming for exactness has enabled Michael and me to uncover important factors that one needs to consider in understanding exchange rate fluctuations. Our models of currency movements have focused on the role of macroeconomic fundamentals and revisions of individual forecasting strategies.

The inclusion of fundamentals would probably surprise our academic colleagues. They have spent decades trying to show that fundamentals matter for currency movements, but have come up empty-handed time and again. Many of them now believe that irrational traders drive currency movements. But this is another odd conclusion. After all, most observers, including economists themselves, blame the dollar’s decline on fundamentals like the U.S. economy’s weakness, the interest-rate differential relative to Europe, and America’s enormous current-account deficit.

The reason macroeconomic fundamentals have disappeared from academic explanations of currency movements is disarmingly simple. Most academic economists treat economics like a form of engineering, limiting themselves to searching for models in which the relationship between the exchange rate and fundamentals is either fixed or changes in mechanical ways.

But in a dynamic world economy, why should anyone believe that a single set of economic fundamentals has mattered in exactly the same way since floating currencies became the norm in the 1970s? IKE acknowledges that finding one true model for currency movements is outside the reach of economic analysis, so it has no problem recognizing that macroeconomic fundamentals really are important for currency movements, which statistical analysis also demonstrates. But which fundamentals are important may change from one point in time to another.

TIE: So IKE does not deliver ready-made forecasting models?

Frydman: Right. If economic theory could deliver an exact forecast of the market price, markets would not play as essential a role as they do. The role of economic theory ends with providing a menu of useful factors and a set of models that generate qualitative predictions of market outcomes. Ultimately, it is up to the forecaster to decide how to use of these factors in predicting future market outcomes.

TIE: What are the implications of your thinking for the subprime mortgage crisis and the role of credit ratings agencies?

Frydman: This is a big topic. Let me just briefly mention two implications.

The rating agencies have generally been better at rating corporate bonds than at rating asset-backed collateralized debt obligations. One reason is that when they rate corporate bonds, they rely on both a mathematical model and their own specialists’ judgment. For subprime-related securities, they had no experience or intuition, not least because the instruments were so new, so they could use only mathematical models.

Beyond ratings, quantitative finance models have been widely used to price various derivative and insurance products. But, for the most part, what is usually referred to as the price of risk has nothing to do with market prices or risk. All that these models can deliver is their creators’ mechanistic view of what markets do. But their supposedly scientific pedigree creates the illusion that all uncertainty can be priced by financial engineers and sold and insured as if it were ordinary probabilistic risk.

Obviously, it’s not surprising that there was such great demand for products that offered to rid financial market decisions of their inherent non-probabilistic uncertainty. But, because contemporary finance models aim to generate exact predictions of outcomes and their likelihoods, they must ignore uncertainty. Unfortunately, this time the consequences of insisting on models that imply exact predictions were not confined to seminar rooms and academic journals.

In an October 19, 2008 op-ed for Financial Times Frydman, Goldberg and Nobel laureate Edmund Phelps discuss their proposed multi-valued rating policy:

…There is plenty of blame to go around. Finance and economics professors devoted their talents to developing abstruse, yet simplistic, models that left out the imperfection of knowledge. Universities have produced two generations of financial engineers who sold the idea that academic models could safely neglect market discovery of risk and prices.

Clearly, regulatory reform must address the ratings agencies’ conflict of interest, given how much their bottom lines have relied on their assessments of structured assets. But doing so will not address their failure to convey the necessarily contingent character of their models and assumptions. The failure to acknowledge this ambiguity obscured substantial risks.

The ratings agencies used statistical models that projected historical default patterns to continue. These patterns showed very low loss rates, thanks to ever rising prices since 1997. With such low loss rates, triple A ratings appeared to be justified. Brave new models tempted the industry to abandon proven prudential procedures, which combined their own judgment and more formal criteria. To be sure, agencies apply stresses to their current procedures. But these stresses are hidden in ratings reports and, as recent events have painfully demonstrated, are woefully inadequate.

And on Nov 6, 2008 there is an excellent Tom Keene interview with Frydman on Bloomberg on the Economy [MP3].