New research recommends using the one-year GDP and the three-month Treasury bill as benchmarks to adjust for distortions across multiple macroeconomic forecasts.
One such comes from experts at Wharton and elsewhere, who have developed a tool to deal with the Achilles Heel of such forecasting: biases rooted in subjective beliefs, such as those for inflation, bank interest rates, and mortgage rates. In a new paper titled “The Subjective Belief Factor,” they identified two forecasts as the benchmarks to adjust for biases across all macroeconomic forecasts: the one-year rate for real GDP growth (adjusted for inflation) and the three-month Treasury bill.
“The main motivation here is that both investors and policymakers rely very heavily on professional forecasts of the macro economy,” said Wharton finance professor Sean Myers, who co-authored the paper with Ricardo De la O, professor of finance and business economics at the University of Southern California’s Marshall School of Business, and Tingyue Cui, a doctoral student in Wharton’s finance department.
Macroeconomic forecasts cover a broad range, including those for recession probabilities, interest rates, and unemployment. Those forecasts are inherently prone to biased expectations, which may not be consistent with historical facts or what the underlying data predicts. Naturally, investors and policymakers “care a lot about potential biases in these forecasts,” Myers said. “But it’s very hard to boil this down into a couple of key features because there are so many different aspects of the macro economy you could consider.”
Myers said their paper is especially relevant in the prevailing environment of macroeconomic uncertainty. He pointed in particular to uncertainty over the economic impact of tariffs, and the “heightened probability” that the U.S. is headed for a recession. “In these times of uncertainty, you have to train and hone your model to be very good at the short-term interest rate and real GDP. You don’t have to focus on trying to address each one of these variables — the tariffs, exports, interest rate changes, or unemployment changes.”
Working Around Distorted Probabilities
Finding a way around the unwieldy task of parsing through numerous variables is the main contribution of the paper. The authors used asset pricing techniques to estimate a Subjective Belief Factor (SBF) to deal with two types of distorted probabilities: one based on subjective expectations, and the other based on preferences or risks. Subjective expectations are at one end of the forecasting spectrum; at the other end are statistical expectations.
The authors developed a single SBF related to real GDP growth and the T-bill rate, which summarizes the differences between subjective and statistical expectations for 24 macroeconomic variables, such as inflation, interest rates, the mortgage rate, and unemployment. It also explains about half the variation in stock returns across 176 anomalies, such as those related to sales, earnings, and profitability at companies; the remaining half of the variation is attributed to preferences or risks.
Myers explained how tools from asset pricing helped in developing the SBF. Factor analysis tools help understand underlying characteristics of stocks, such as book-to-market, or leverage. “We say we don’t have 1,001 unique stock phenomena, but there are a handful of key factors that we want to understand, which we think can summarize variations across many different stocks.”
Tingyue Cui described the SBF tool as “a framework” to summarize people’s expectations as reported in the Survey of Consumer Expectations, a quarterly survey of U.S. macroeconomic forecasts. Adjusting for biases would help consumers more accurately make household decisions, where they may be overstating the likelihood of a house price increase, or understating the probability of unemployment, Myers added.
Why GDP and the 3-month Interest Rate Are Key
The SBF is trained to focus on the forecasts of real GDP growth and the three-month T-bill interest rate for good reason. “Once you understand biases in those two forecasts, you can essentially replicate and predict the biases in all the other forecasts,” Myers said. “So, you don’t have to go study the six-month interest rate forecast, or the unemployment forecast, or the exports forecast. If you can just get a good model of the bias in the GDP forecast and the three-month T-bill forecast, you can then extrapolate all the other biases off of those two. Just specialize in training your statistical or machine learning model on these two underlying factors, and that will give you most of the reward.”
Myers offered a couple of examples of how a macroeconomic forecast could be linked to those for real GDP growth and the three-month T-bill rate. A forecast on unemployment, for instance, could be adjusted for the historical relationship between GDP and unemployment. If the historical trend shows that a one-percentage-point higher GDP is generally linked to half-a-percentage-point lower unemployment, one could broadly conclude that high GDP corresponds with low unemployment.
Similarly, in order to adjust for bias in an inflation forecast, one could study historical relationships between inflation on the one hand, and the short-term interest rate and GDP on the other. “You could use that to guess how biased that implies the inflation forecast would be, given that high interest rates are generally associated with high inflation. Or, since inflation is generally pro-cyclical, it tends to be high when GDP is high,” Myers explained. “The stronger that historical relationship is, the better these predictions are.”
How Biases Distort Forecasts
Myers offered two instances of how biases lead to inaccurate forecasts. The most recent instance of biased expectations relates to interest-rate forecasts in 2022, when the economy was recovering from the COVID-era downturn. Back then, interest rates had been pegged at near-zero levels as part of the stimulus injections to lift the economy.
Contrary to forecasts, interest rates began climbing after early 2022 to about 5% in the next 15 months. “It seems like forecasters were understating the probability of an interest rate hike in 2022,” Myers said. “But historical statistical models would have told you that with inflation that high and unemployment at that level, they should have been anticipating an interest rate hike.”
An earlier period when forecasters got it wrong was in the early 1980s, when they assumed that the super-high inflation of the mid to late 1970s would persist. “Long-term bond yields were still very high, essentially predicting that inflation would stay high, Myers said. “People seem to underestimate the probability that the Fed could actually bring down inflation.”
But the then Federal Reserve chairman Paul Volcker eventually raised interest rates to around 20% in 1981, taming back inflation to around 2% by the mid-1980s.
Who Benefits From the SBF, and How
Myers listed the major takeaways from the SBF tool for various constituencies:
First off, forecasters can fix biases in dozens of macroeconomic forecasts by just addressing a couple of key underlying issues.
Traders or investors can make better decisions by adjusting forecasts for biases. Early adapters could have an advantage over the rest of the market, but widespread adoption will make the market as a whole more efficient.
More accurate forecasting could help make market rates more realistic. For instance, in the 1980s, when long-term bond yields were excessively high on flawed expectations of continued high inflation, the bond markets understated the probability of interest rate increases. Consequently, companies paid unreasonably high interest rates on borrowings.
Accurate forecasts also make way for more efficient allocation of capital across companies. In that setting, for instance, companies will not get loans at excessively low or excessively high interest rates.
Individual consumers may not be aware of how biases or forecasting errors affect them, but more accurate forecasts could mean that they don’t overpay for their credit card debt or mortgage. “It would impact the quality of life for the person on the street by simply improving the efficiency of the interest rates that they’re being offered,” said Myers.
Where the SBF May Fall Short
“If you care about macroeconomic variables, the one-year real GDP rate and the short-term interest rate appear to be the two key factors in adjusting for biases,” Myers said. “But if you step outside that context, they may not be very good at summarizing things.”
For instance, the SBF tool won’t work as well in cases where the GDP and the T-bill rates are not tightly related to all the other variables being studied. For example, forecasts of housing starts are only “moderately related” to real GDP and the three-month interest rate, Myers noted. Housing starts are also related to many elements that are specific to housing, because of which the SBF isn’t very good at predicting biases. The SBF isn’t very useful in addressing biases in state and local finances, or in narrow segments such as service sector hospitality prices, he added.
“Anytime you’re in a scenario where there are dozens of variables which you need to understand, your first instinct should not be to study them one by one,” Myers said. “Your first instinct should always be to condense. Try to boil them down to a couple of key characteristics.”
