Economic+forecasting

**What is a Forecast?**
A forecast is any statement abut the future. Such statements may be well founded, or simply lack any sound basis; they are capable of being accurate or inaccurate on any given occasion, or on average; precise or imprecise. Forecasts are produced by methods as diverse as well-tests systems of hundreds of econometrically-estimated equations, through to methods which have scarcely any observable basis (such as the forecasts made by the 2009 Derby winer made on the 31st of December 2000 - before the entrants are even known.) Thus, forecasting is potentially a vast subject. Historically, almost every conceivable method has been tried, with the legacy that there are in excess of 36 different words in English for the activity of "foretelling" in a broad sense what the future might bring forth.


 * What is Economic Forecasting?**

==== Economic Forecasting is the process of attempting to predict the future condition of the economy. This involves the use of statistical models utilizing variables sometimes called indicators. Some of the most well-known economic indicators include inflation and interest rates, GDP growth/decline, retail sales and unemployment rates. ====

What can be forecast? Since it is merely a statement about the future, anything can be forecast, ranging from next month’s rate of consumer price inﬂation, tomorrow’s weather patterns, the average rise in sea levels by the end of the third millennium, through the earth’s population at the same date, to the value of the Dow Jones index at the start of 2010. We are not claiming that the resulting forecasts are necessarily useful in any sense: consider, for example, a forecast that the ﬁrst Extra Terrestrial to land on Earth will be six meters tall, blue, and will arrive in New York on July 4th, 2276 to celebrate the quincentenary of the U.S.A. Even if such a claim were to prove correct, it would be of no value for the next 250 years; and of course, it is anyway essentially certain to be incorrect.

How conﬁdent can we be in forecasts? Clearly, our conﬁdence will depend on how well-based the forecasts are. Mere guesses should not inspire great conﬁdence; forecasts from well-tested approaches  may be viewed more hopefully. Unfortunately, even the latter is not enough. The trouble is that the future is uncertain. There are two distinct senses in which this applies, expressed by Maxine Singer in her “Thoughts of a Nonmillenarian” (Bulletin of the American Academy of Arts and Sciences, 1997, 51, 2, p. 39) as: Because of the things we don’t know [that] we don’t know, the future is largely unpredictable. But some developments can be anticipated, or at least imagined, on the basis of existing knowledge.

 Little can be done in advance about uncertainty stemming from “things we don’t know we don’t know.” However, the apparent randomness of outcomes within the realms we do understand, which we will call “measurable uncertainty,” can  often be usefully communicated to the user of a forecast. This usually takes the form of a forecast interval around a “point” forecast, the latter then being viewed as the central tendency, or “most likely” outcome. For example, the statement  that “the moon is exactly 5,000 miles away” is very precise (but wholly inaccurate), and taken literally would be associated with a forecast interval of length zero. On the other hand, the statement that “the moon lies between 1,000 and 1  billion miles away” is correct, but very imprecise, having a huge forecast interval. More sophisticated presentations of measurable uncertainty include density forecasts; namely, estimates of the probability distribution of the possible future  outcomes. The Bank of England tries to present its Inﬂation Report forecasts in this last form, using a “fan chart” where uncertainty fans out into the future in ever wider bands of lighter color (unfortunately, they chose red for the inﬂation forecasts and green for output, so these were called “rivers of blood” and “rivers of bile” respectively.     How is forecasting done generally?    There are many ways of making forecasts. These include formal model-based statistical analyses, statistical analyses not based on parametric models, in- formal “back-of-the-envelope” calculations, simple extrapolations, “leading indic-  ators,” “chartist” approaches, “informed judgment,” tossing a coin, guessing, and “hunches.” It is difﬁcult to judge the frequency with which each of these methods is used in practice, but most occur regularly in our everyday lives. In   earlier times, tea leaves, entrails, movements of the stars, etc., all were tried – without great success so far as we can ascertain – but some (such as astrology) remain in use today. This book, for better or worse, will focus on formalstatistical  approaches.   How is forecasting done by economists?  In economics, methods of forecasting include:  1 guessing, “rules of thumb,” or “informal models”;  2 expert judgment; <span style="mso-layout-grid-align: none; mso-pagination: none; tab-stops: 28.0pt 56.0pt 84.0pt 112.0pt 140.0pt 168.0pt 196.0pt 224.0pt 252.0pt 280.0pt 308.0pt 336.0pt; text-autospace: none;"> 3 extrapolation; <span style="mso-layout-grid-align: none; mso-pagination: none; tab-stops: 28.0pt 56.0pt 84.0pt 112.0pt 140.0pt 168.0pt 196.0pt 224.0pt 252.0pt 280.0pt 308.0pt 336.0pt; text-autospace: none;"> 4 leading indicators; <span style="mso-layout-grid-align: none; mso-pagination: none; tab-stops: 28.0pt 56.0pt 84.0pt 112.0pt 140.0pt 168.0pt 196.0pt 224.0pt 252.0pt 280.0pt 308.0pt 336.0pt; text-autospace: none;"> 5 surveys; <span style="mso-layout-grid-align: none; mso-pagination: none; tab-stops: 28.0pt 56.0pt 84.0pt 112.0pt 140.0pt 168.0pt 196.0pt 224.0pt 252.0pt 280.0pt 308.0pt 336.0pt; text-autospace: none;"> 6 time-series models; and <span style="mso-layout-grid-align: none; mso-pagination: none; tab-stops: 28.0pt 56.0pt 84.0pt 112.0pt 140.0pt 168.0pt 196.0pt 224.0pt 252.0pt 280.0pt 308.0pt 336.0pt; text-autospace: none;"> 7 econometric systems. <span style="mso-layout-grid-align: none; mso-pagination: none; tab-stops: 28.0pt 56.0pt 84.0pt 112.0pt 140.0pt 168.0pt 196.0pt 224.0pt 252.0pt 280.0pt 308.0pt 336.0pt; text-autospace: none;"> Guessing and related methods only rely on luck. While that may be a minimal assumption compared to the other methods we will discuss, guessing is not generally a useful method: “good” guesses are often reported, and bad ones quietly ignored; and the uncertainty attaching to each guess is usually impossible to evaluate in advance. If many individuals guess, some will be “right” by chance, but that hardly justiﬁes the approach (otherwise economists will start producing thousands of forecasts and claiming success whenever any one of them is accurate). Expert judgment is usually part of a forecasting approach, but lacks validation when it is the sole component, even if at any point in time, some “oracle” manages to have forecasted accurately. Unfortunately, systematic success proves elusive even to experts, and no one can predict which oracle will be successful next (note the recent advice to ignore past performance when choosing a  <span style="mso-layout-grid-align: none; mso-pagination: none; tab-stops: 28.0pt 56.0pt 84.0pt 112.0pt 140.0pt 168.0pt 196.0pt 224.0pt 252.0pt 280.0pt 308.0pt 336.0pt; text-autospace: none;"> fund manager!). Extrapolation is ﬁne so long as the tendencies persist, but that is itself doubtful: the telling feature is that different extrapolators are used at different points in  <span style="mso-layout-grid-align: none; mso-pagination: none; tab-stops: 28.0pt 56.0pt 84.0pt 112.0pt 140.0pt 168.0pt 196.0pt 224.0pt 252.0pt 280.0pt 308.0pt 336.0pt; text-autospace: none;"> time. Moreover, forecasts are most useful when they predict changes in tendencies, which extrapolative methods are likely to miss. Many a person has bought shares, or a house, at the peak of a boom.... Forecasting based on leading indicators requires a stable relationship between the variables that “lead” and the variables that are “led.” When the reasons for <span style="mso-layout-grid-align: none; mso-pagination: none; tab-stops: 28.0pt 56.0pt 84.0pt 112.0pt 140.0pt 168.0pt 196.0pt 224.0pt 252.0pt 280.0pt 308.0pt 336.0pt; text-autospace: none;"> the lead are clear, as with orders preceding production, then the indicators may be useful, but otherwise are liable to give misleading information. Even for such “obvious” leading indicators as housing starts leading to completed dwellings, <span style="mso-layout-grid-align: none; mso-pagination: none; tab-stops: 28.0pt 56.0pt 84.0pt 112.0pt 140.0pt 168.0pt 196.0pt 224.0pt 252.0pt 280.0pt 308.0pt 336.0pt; text-autospace: none;"> the record is poor (because the delay can narrow and widen dramatically in housing market booms and busts – or with very severe weather). Surveys of consumers and businesses can be informative about future events, <span style="mso-layout-grid-align: none; mso-pagination: none; tab-stops: 28.0pt 56.0pt 84.0pt 112.0pt 140.0pt 168.0pt 196.0pt 224.0pt 252.0pt 280.0pt 308.0pt 336.0pt; text-autospace: none;"> but rely on plans being realized. Again we see “many a slip twixt cup and lip”: adverse changes in the business “climate” can induce radical revisions to plans, since it is less costly to revise a plan than the actuality. Time-series models which describe the historical patterns of data are popular fore- <span style="mso-layout-grid-align: none; mso-pagination: none; tab-stops: 28.0pt 56.0pt 84.0pt 112.0pt 140.0pt 168.0pt 196.0pt 224.0pt 252.0pt 280.0pt 308.0pt 336.0pt; text-autospace: none;"> casting methods, and have often been found to be competitive relative to econometric systems of equations (particularly in their multivariate forms). These are the work-horse of the forecasting industry, and several chapters below explainand <span style="mso-layout-grid-align: none; mso-pagination: none; tab-stops: 28.0pt 56.0pt 84.0pt 112.0pt 140.0pt 168.0pt 196.0pt 224.0pt 252.0pt 280.0pt 308.0pt 336.0pt; text-autospace: none;"> analyze variants thereof. But like all other methods, they focus on “measurable uncertainty.” Econometric systems of equations are the main tool in economic forecasting. These <span style="mso-layout-grid-align: none; mso-pagination: none; tab-stops: 28.0pt 56.0pt 84.0pt 112.0pt 140.0pt 168.0pt 196.0pt 224.0pt 252.0pt 280.0pt 308.0pt 336.0pt; text-autospace: none;"> comprise equations which seek to “model” the behavior of discernible groups of economic agents (consumers, producers, workers, investors, etc.) assuming a considerable degree of rationality – moderated by historical experience. The <span style="mso-layout-grid-align: none; mso-pagination: none; tab-stops: 28.0pt 56.0pt 84.0pt 112.0pt 140.0pt 168.0pt 196.0pt 224.0pt 252.0pt 280.0pt 308.0pt 336.0pt; text-autospace: none;"> advantages to economists of using formal econometric systems of national economies are to consolidate existing empirical and theoretical knowledge of how economies function, provide a framework for a progressive research strategy leading to increased understanding over time, help to explain their own failures, as well as provide forecasts and policy advice. Econometric and time-series models <span style="mso-layout-grid-align: none; mso-pagination: none; tab-stops: 28.0pt 56.0pt 84.0pt 112.0pt 140.0pt 168.0pt 196.0pt 224.0pt 252.0pt 280.0pt 308.0pt 336.0pt; text-autospace: none;"> are the primary methods of forecasting in economics, but “judgment,” “indicators,” and even “guesses” may modify the resulting forecasts. <span style="mso-layout-grid-align: none; mso-pagination: none; tab-stops: 28.0pt 56.0pt 84.0pt 112.0pt 140.0pt 168.0pt 196.0pt 224.0pt 252.0pt 280.0pt 308.0pt 336.0pt; text-autospace: none;"> <span style="mso-layout-grid-align: none; mso-pagination: none; tab-stops: 28.0pt 56.0pt 84.0pt 112.0pt 140.0pt 168.0pt 196.0pt 224.0pt 252.0pt 280.0pt 308.0pt 336.0pt; text-autospace: none;"> How can one measure the success or failure of forecasts? <span style="mso-layout-grid-align: none; mso-pagination: none; tab-stops: 28.0pt 56.0pt 84.0pt 112.0pt 140.0pt 168.0pt 196.0pt 224.0pt 252.0pt 280.0pt 308.0pt 336.0pt; text-autospace: none;"> A forecast might reasonably be judged “successful” if it was close to the outcome, but that judgment depends on how “close” is measured. Reconsider our example of “guessing” the distance to the moon: it is apparent that accuracy and precision are two dimensions along which forecasts may be judged. To the layman, a very precise forecast that is highly inaccurate might be thought undesirable, as might an accurate but very imprecise forecast: and experts concur – the “goldstandard” is an accurate and precise forecast. Failure is easier to discern: a forecast is a failure if it is inaccurate by an amount that is large relative to its claimed precision. Thus, forecasters are squeezed between wanting accurate and precise forecasts, yet not claiming so much precision that they regularly fail. The notion of “unbiasedness,” whereby forecasts are centered on outcomes, is used in technical analyses to measure accuracy; whereas that of small variance, so only a narrow range of outcomes is compatible with the forecast statement, measures precision.