month of that quarter – after the quarter was almost over. In the current decade, such projections were 0.8 percent off from what was reported. (Note: This is judged by the mean absolute error – the absolute magnitude of an error without regard to whether the forecast was too high or too low.) Moreover, these “last minute” projections were even farther off in earlier decades.
Figure 1.2 In the Final Month of a Quarter, Forecasters' Growth Forecasts for That Quarter Can Still Err Substantially
Source: Federal Reserve Bank of Philadelphia.
Moving forward, we discuss how the various economic “weather reports” can suggest winter and summer on the same day! Let's note, too, that some of the key indicators of tomorrow's business weather are subject to substantial revisions. At times it seems like there are no reliable witnesses, because they all change their testimony under oath. In later chapters we discuss how to address these challenges.
2. History does not always repeat or even rhyme.
Forecasters address the future largely by extrapolating from the past. Consequently, prognosticators can't help but be historians. And just as the signals on current events are frequently mixed and may be subject to revision, so, too, when discussing a business or an economy, are interpretations of prior events. In subsequent chapters, we discuss how to sift through history and judge what really happened – a key step in predicting, successfully, what will happen in the future.
The initially widely acclaimed book, This Time Is Different: Eight Centuries of Financial Follies by Carmen Reinhart and Kenneth Rogoff, provides a good example of the difficulties in interpreting history in order to give advice about the future.12 Published in 2011, the book first attracted attention from global policymakers with its conclusion that, since World War II, economic growth turned negative when the government debt/GDP ratio exceeded 90 percent. Two years later, other researchers discovered calculation errors in the authors' statistical summary of economic history. Looking for repetitive historical patterns can be tricky!
3. Statistical crosscurrents make it hard to find safe footing.
Even if the past and present are clear, divining the future remains challenging when potential causal variables (e.g., the money supply and the Federal purchases of goods and services) are headed in opposite directions. However, successful and influential forecasters must avoid being hapless “two-handed economists” (i.e., “on the one hand, but on the other hand”).
Moreover, one's statistical coursework at the college and graduate level does not necessarily solve the problem of what matters most when signals diverge. Yes, there are multiple regression software packages readily available that can crank out estimated regression (i.e., response) coefficients for independent causal variables. But, alas, even the more advanced statistical courses and textbooks have yet to satisfactorily surmount the multicollinearity problem. That is when two highly correlated independent variables “compete” to claim historical credit for explaining dependent variables that must be forecast. As a professional forecaster, I have not solved this problem but have been coping with it almost every day for decades. As we proceed, you will find some helpful tips on dealing with this challenge.
4. Behavioral sciences are inevitably limited.
There have been quantum leaps in the science of public opinion polling since the fiasco of 1948, when President Truman's reelection stunned pollsters. Nevertheless, there continue to be plenty of surprises (“upsets”) on election night. Are there innate limits to humans' ability to understand and predict the behavior of other humans? That was what the well-known conservative economist Henry Hazlitt observed in reaction to all of the hand wringing about “scientific polling” in the aftermath of the 1948 debacle. Writing in the November 22, 1948, issue of Newsweek, Hazlitt noted: “The economic future, like the political future, will be determined by future human behavior and decisions. That is why it is uncertain. And in spite of the enormous and constantly growing literature on business cycles, business forecasting will never, any more than opinion polls, become an exact science.”13
In other words, forecast success or failure can reflect “what we don't know that we don't know” (generalized uncertainty) more than “what we know” (risk).
5. The most important determinants may not be measureable.
Statistics are all about measurement. But what if you cannot measure what matters? Statisticians often approach this stumbling block with a dummy variable. It is assigned a zero or one in each examined historical period (year, quarter, month, or week) according to whether the statistician believes that the unmeasurable variable was active or dormant in that period. (For example, when explaining U.S. inflation history with a regression model, a dummy variable might be used to identify periods when there were price controls.) If the dummy variable in an estimated multiple regression equation achieves statistical significance, the statistician can then claim that it reflects the influence of the unmeasured, hypothesized causal factor.
The problem, though, is that a statistically significant dummy variable can be credited for anything that cannot be otherwise accounted for. The label attached to the dummy variable may not be a true causal factor useful in forecasting. In other words, there can be a naming contest for a dummy variable that is statistically sweeping up what other variables cannot explain. There are some common sense approaches to addressing this problem, and we discuss them later.
6. There can be conflicts between the goal of accuracy and the goal of pleasing a forecaster's everyday workplace environment.
Many of the most publicly visible and influential forecasters – especially securities analysts and investment bank economists – have job-related considerations that can influence their advice about the future. It is ironic that financial analysts and economists whose good work has earned them national recognition can find pressures at the top that complicate their ability to give good advice once the internal and external audience enlarges.
Many Wall Street economists, for instance, are employed by fixed-income or currency trading desks. Huge amounts of their firms' and their clients' money are positioned before key economic statistics are reported. This knowledge might understandably make a forecaster reluctant to go against the consensus. And, as we discuss shortly, there can be other work-related pressures not to go against the grain as well.
Are trading desks' economists' forecasts sometimes made to assist their employers' business?
It is hard, if not impossible, to gauge how much and how frequently forecasts are conditioned by an employer's business interests. However, it can be observed that certain types of behavior are consistent with the hypothesis that forecasts are being affected in this manner. For instance, the economist Takatoshi Ito at the University of Tokyo has authored research suggesting that foreign exchange rate projections are systematically biased toward scenarios that would benefit the forecaster's employer. He has attached the label “wishful expectations” to such forecasts.14
What is the effect of the sell-side working environment on stock analysts' performance?
In order to be successful, sell-side securities analysts at brokerage houses and investment banks must, in addition to performing their analytical research, spend time and effort marketing their research to their firms' clients. In buy-side organizations, such as pension funds, mutual funds, and hedge funds, analysts generally do not have these marketing responsibilities. Do the two different work environments make a difference in performance?
Henry Hazlitt, “Pitfalls of Forecasting,”
Takatoshi Ito, “Foreign Exchange Rate Expectations: Micro Survey Data,”
