There are many good books and articles on time-series econometrics, which cover forecasting, as well as books and handbooks specifically on forecasting. A highly selective set of examples of the former include Harvey (1990), Hamilton (1994), and Hendry (1995), and of the latter, Theil (1958, 1971), Box and Jenkins (1970), Klein (1971), Granger and Newbold (1977), and Clements and Hendry (1998, 1999), including various handbooks, Clements and Hendry (2002), Elliott et al. (2006), Clements and Hendry (2011), and Elliott and Timmermann (2013). There are also good popular science books attempting to explain forecasting to the general reader: for example, Silver (2012), Tetlock and Gardener (2015), Goodwin (2017), Castle et al. (forthcoming).
Why then a text specifically on survey expectations? Survey expectations require a separate treatment and are not covered in the works on time-series econometrics and forecasting. The main reason is that the literature on time-series econometrics and forecasting for the most part considers forecasts from clearly articulated models or methods. The properties of these models and methods can be analysed mathematically once we have assumed a data generating process. For example, Clements and Hendry (1999) provide a thorough treatment of forecasting with vector equilibrium (‘error’) correction models when the data generating process is a cointegrated system of integrated variables subject to location shifts. Or alternatively, if the mathematics is intractable, the forecasts from the models or methods can be analysed by Monte Carlo simulation. Finally, of course, the performance of the models or methods can be assessed by running the models on historical data for different epochs, and the impact of changing the model selection, specification, or estimation strategies can be calculated.
Survey expectations by contrast are not based on models or methods that are typically known to anybody other than the forecaster.1 It seems unlikely that a forecaster uses a given model or method in an unadulterated way for any length of time, that is, without subjective adjustments to the forecasts. If so, then even in principle a sequence of forecasts made over a period of time does not represent the forecast performance of a given model or method. The bottom line is that survey forecasts cannot be evaluated as the forecasts from a specific model (or method). Hence the survey forecasts stand alone and are neither supported by, nor brought into question by, association with a given modelling approach. This alone means that analyses which regard forecasting as an extension of time-series econometrics will not be directly relevant for survey expectations.
In this book we consider the particular challenges to, and potential rewards from, studying survey expectations. Chapter 2 describes the nature of the survey data. In Chap. 3 we address some of the difficulties posed by the way in which survey expectations are elicited and, in particular, the presentation of what the econometrician would like to regard as probability distributions in the form of histograms. As we will see, problems arise throughout the book in terms of how to interpret the survey responses in terms of first moments and second moments, and we explain how these challenges are dealt with. Chapters 4 and 5 consider the evaluation of the survey point predictions and the probability distributions, respectively. Chapter 6 considers the consistency of an individual’s point forecasts and probability distributions and explores one possible explanation for the apparent discrepancies—that the probability forecasts have been rounded.
We then turn our attention to the reasons why we might be interested in survey expectations. One of the traditional reasons to be interested in forecasting is that out-of-sample forecast performance is often regarded as the gold standard for evaluating a model. This is driven by concerns that the in-sample fit of the model to the data can be manipulated by the researcher who undertakes ‘searches for significance’ across large numbers of potential explanatory variables, and the final model may reflect chance relationships specific to the sample of data under study. Out-of-sample performance ought to give a truer picture of the explanatory power of the model and, therefore, the argument goes, of the validity of the economic theory that underpins the model’s set of explanatory variables. That there is a simple mapping between out-of-sample forecast performance and the validation of the theory encapsulated in the economic model has been challenged—see, for example, Clements and Hendry (2005).
In any case, although this possible justification for an interest in forecasting does not extend to survey expectations, survey expectations have much to offer. Firstly, survey expectations in principle draw on a wide variety of information and assimilate information from various sources, some of which might not be easily codified in a formal model. Surveys might be expected to be more accurate than model forecasts, at least at short horizons, to the extent that they are based on an up-to-date reading of the current state of the economy, tempered by judgement. Secondly, survey expectations in the form of histograms provide direct estimates of uncertainty, at least relative to theoretically unsatisfactory proxies such as disagreement between agents (see Chap. 7). Thirdly, from a behavioural perspective, they offer insights into how economic agents form their expectations. There are theories of how agents behave, and how they ought to behave, and survey expectations offer the prospect of discovering which of these are supported by actual, directly reported expectations. Theories of expectations formation are discussed in Chap. 8. Fourthly, changes in sentiment or expectations have long been accorded a role in explaining business-cycle variation. Survey expectations have been included in the popular structural vector autoregressive (SVAR) models as a source of expectations, enabling the effects of exogenous changes in expectations to be calculated. Relatedly, survey expectations have been used to solve the problem of non-fundamentalness in SVARs, whereby the econometrician’s information set would otherwise be less than that of the agents in the economy (see Chap. 9).
In writing this book I have drawn on published material which I have either authored or co-authored. Section 4.5 draws on Clements, M.P. (2014), US Inflation Expectations and Heterogeneous Loss Functions, 1968–2010, Journal of Forecasting, 33(1), 1–14. Material from this article is used with the permission of John Wiley and Sons. Chapter 5 is based on Clements, M.P. (2018), Are Macroeconomic Density Forecasts Informative?, International Journal of Forecasting, 34(2), 181–198. Material from this article is used with the permission of Elsevier. Sections 6.1 and 6.2 are based on Clements, M.P. (2014), Probability Distributions or Point Predictions? Survey Forecasts of US Output Growth and Inflation, International Journal of Forecasting, 30(1), 99–117. Material from this article is used with the permission of Elsevier. The remainder of Chap. 6 is based on Clements, M.P., (2011), An Empirical Investigation of the Effects of Rounding on the SPF Probabilities of Decline and Output Growth Histograms, Journal of Money, Credit and Banking, 43(1), 207–220. Material from this article is used with the permission of John Wiley and Sons. Finally, Chap. 7 is based on joint work with Ana B. Galvão: Clements, M.P. and Galv ão, A.B. (2017), Model and Survey Estimates of the Term Structure of US Macroeconomic Uncertainty, International Journal of Forecasting 33(3), 591–604. Material from this article is used with the permission of Elsevier.
References
- Batchelor, R., & Dua, P. (1991). Blue Chip rationality tests. Journal of Money, Credit and Banking, 23, 692–705.
- Box, G. E. P., & Jenkins, G. M. (1970). Time series analysis, forecasting and control. San Francisco, CA: Holden-Day.
- Castle, J. L., Clements, M. P., & Hendry, D. F. (forthcoming). Forecasting: An essential guide. Yale University Press.
- Clements, M. P., & Hendry, D. F. (1998). Forecasting economic time series. Cambridge: Cambridge University Press. The Marshall lectures on economic forecasting.
- Clements, M. P., & Hendry, D. F. (1999). Forecasting non-stationary economic time series. Cambridge, MA: MIT Press.
- Clements, M. P., & Hendry, D. F. (Eds.). (2002). A companion to economic forecasting. Oxford: Blackwells.
- Clements,...