Search

Parallel Sessions F: Real Time Data

The recent literature has suggested that macroeconomic forecasters may have asymmetric loss functions, and that there may be heterogeneity across forecasters in the degree to which they weigh under and over-predictions. Using an individual-level analysis that exploits the SPF respondentsí histogram forecasts, we find little evidence of asymmetric loss for the inflation forecasters.

Survey respondents who make point predictions and histogram forecasts of macrovariables reveal both how uncertain they believe the future to be, ex ante, as well as their ex post performance. Macroeconomic forecasters tend to be overconÖdent at horizons of a year or more, but over-estimate the uncertainty surrounding their predictions at short horizons.

We consider whether survey respondentsíprobability distributions, reported as histograms, provide reliable and coherent point predictions, when viewed through the lens of a Bayesian learning model, and whether they are well calibrated more generally. We argue that a role remains for eliciting directly-reported point predictions in surveys of professional forecasters.

We present a novel approach to assessing the attentiveness of professional forecasters to news about the macroeconomy. We find evidence that professional forecasters, taken as a group, do not always update their estimates of the current state of the economy to reflect the latest releases of revised estimates of key data.

Real-time estimates of output gaps and inflation trends differ from the values that are obtained using data available long after the event. Part of the problem is that the data on which the real-time estimates are based is subsequently revised. We show that vector-autoregressive models of data vintages provide forecasts of post-revision values of future observations and of already-released observations capable of improving real-time output gap and inflation trend estimates. Our findings indicate that annual revisions to output and inflation data are in part predictable based on their past vintages.

We investigate two characteristics of survey forecasts that are shown to contribute to their superiority over purely model-based forecasts. These are that the consensus forecasts incorporate the effects of perceived changes in the long-run outlook, as well as embodying departures from\ud the path toward the long-run expectation. Both characteristics on average tend to enhance forecast accuracy. At the level of the individual forecasts, there is scant evidence that the second characteristic enhances forecast accuracy, and the average accuracy of the individual forecasts can be improved by applying a mechanical correction.

We show how to improve the accuracy of real-time forecasts from models that include autoregressive terms by estimating the models on 'lightly-revised' data instead of using data from the latest-available vintage. Forecast accuracy is improved by reorganizing the data vintages employed in the estimation of the model in such a way that the vintages used in estimation are of a similar maturity to the data in the forecast loss function. The size of the expected reductions in mean squared error depend on the characteristics of the data revision process.\ud Empirically, we find RMSFE gains of 2-4% when forecasting output growth and inflation with AR models, and gains of the order of 8% with ADL models.

We consider the possibility that respondents to the Survey of Professional Forecasters round their probability forecasts of the event that real output will decline in the future. We make various assumptions about how forecasters round their forecasts, including that individuals have constant patterns of responses across forecasts. Our primary interests are the impact of rounding on assessments of the internal consistency of the probability forecasts of a decline in real output and the histograms for annual real output growth, and on the relationship between the probability forecasts and the point forecasts of quarterly output growth.

A comparison of the point forecasts and the central tendencies of probability distributions of inflation and output growth of the SPF indicates that the point forecasts are sometimes optimistic relative to the probability distributions. We consider and evaluate a number of possible explanations for this finding, including the degree of uncertainty concerning the future, computational costs, delayed updating, and asymmetric loss. We also consider the relative accuracy of the two sets of forecasts.

The recent literature suggests that Örst announcements of real output growth in the US have predictive power for the future course of the economy. We show that this need not point to a behavioural relationship, whereby agents respond to the announcement, but may instead simply be a by-product of the data revision process. Initial estimates are subsequently subject to a number of rounds of revisions: the nature of these revisions is shown to be key in determining any apparent relationship between Örst announcements and the future course of the economy.

Many macroeconomic series such as US real output growth are sampled quarterly, although potentially useful predictors are often observed at a higher frequency. We look at whether a mixed data-frequency sampling (MIDAS) approach can improve forecasts of output growth. The MIDAS approach is compared to other ways of making use of monthly data to predict quarterly output growth. The MIDAS specification used in the comparison employs a novel way of including an autoregressive term. We find that the use of monthly data on the current quarter leads to significant improvement in forecasting current and next quarter output growth, and that MIDAS is an effective way of exploiting monthly data compared to alternative methods. We also exploit the best method to use the monthly vintages of the indicators for real-time forecasting.

We ask whether the different types of forecasts made by individual survey respondents are mutually consistent, using the SPF survey data. We compare the point forecasts and central tendencies of probability distributions matched by individual respondent, and compare the forecast probabilities of declines in output with the probabilities implied by the probability distributions. When the expected associations between these different types of forecasts do not hold for some individuals, we consider whether the discrepancies we observe are consistent with rational behaviour by agents with asymmetric loss functions.

We consider tests of forecast encompassing for probability forecasts, for both quadratic and logarithmic scoring rules. We propose test statistics for the null of forecast encompassing, present the limiting distributions of the test statistics, and investigate the impact of estimating the forecasting models’ parameters on these distributions. The small-sample performance of the various statistics is investigated, both in terms of small numbers of forecasts and model estimation sample sizes. Two empirical applications show the usefulness of the tests for the evaluation of recession probability forecasts from logit models with different leading indicators as explanatory variables, and for evaluating survey-based probability forecasts.

Although many macroeconomic series such as US real output growth are sampled quarterly, many potentially useful predictors are observed at a higher frequency. We look at whether a recently developed mixed data-frequency sampling (MIDAS) approach can improve forecasts of output growth and inflation. We carry out a number of related real-time forecast comparisons using various indicators as explanatory variables. We find that MIDAS model forecasts of output growth are more accurate at horizons less than one quarter using coincident indicators ; that MIDAS models are an effective way of combining information from multiple indicators ; and that the forecast accuracy of the unemployment-rate Phillips curve for inflation is enhanced using the MIDAS approach.

We ask whether the di§erent types of forecasts made by individual survey respondents are mutually consistent, using the SPF survey data. We compare the point forecasts and central tendencies of probability distributions matched by individual respondent, and compare the forecast probabilities of declines in output with the probabilities implied by the probability distributions. When the expected associations between these di§erent types of forecasts do not hold for some idividuals, we consider whether the discrepancies we observe are consistent with rational behaviour by agents with asymmetric loss functions.

We evaluate the forecasting performance of a number of systems models of US shortand long-term interest rates. Non-linearities, induding asymmetries in the adjustment to equilibrium, are shown to result in more accurate short horizon forecasts. We find that both long and short rates respond to disequilibria in the spread in certain circumstances, which would not be evident from linear representations or from single-equation analyses of the short-term interest rate.

We propose testing for business cycle asymmetries in Markov-switching autoregressive (MS-AR) models. We derive the parametric restrictions on MS-AR models that rule out types of asymmetries such as deepness, steepness, and sharpness, and set out a testing procedure based on Wald statistics which have standard asymptotics. For a two-regime model, such as that popularised by Hamilton (1989), we show that deepness implies sharpness (and vice versa) while the process is always non-steep. We illustrate with two and three-state MS models of US GNP growth, and with models of US output and employment. Our findings are compared with those obtained from standard non-parametric tests.

In this paper we investigate the multi-period forecast performance of a number of empirical selfexciting threshold autoregressive (SETAR) models that have been proposed in the literature formodelling exchange rates and GNP, amongst other variables. We take each of the empirical SETAR models in turn as the DGP to ensure that the ‘non-linearity’ characterises the future, and compare the forecast performance of SETAR and linear autoregressive models on a number of quantitative and qualitative criteria. Our results indicate that non-linear models have an edge in certain states of nature but not in others, and that this can be highlighted by evaluating forecasts conditional upon the regime.