Abstract
This paper considers the problem of assessing the distributional properties (normality and symmetry) of macroeconomic forecast errors of G7 countries for the purpose of fan-chart modelling. Our results indicate that the assumption of symmetry of the marginal distribution of forecast errors is reasonable, whereas the assumption of normality is not, making symmetric prediction intervals clearly preferable.
Similar content being viewed by others
Notes
Hammond et al. (2012) survey the (inflation) reports of 27 central banks, out of which 20 banks provide prediction intervals officially.
The only exception, the author is aware of, is Reifschneider and Tulip (2019) where the appropriate Monte Carlo critical values are used when testing for normality of the US forecast errors.
It is only fair to note that the authors are aware of this shortcoming (see Reifschneider and Tulip 2007, pp. 19–20).
A MATLAB code is available to researchers upon request from the author.
Note that the null hypothesis can be alternatively stated as: \(\mathcal {H}_{0}^N: F = N(0,\sigma ^2)\) since the forecast errors should be zero-mean stochastic processes. However, empirical evidence suggests that the forecast errors are biased in small samples. The official forecasts are thus corrected for historically observed biases in forecast errors. Therefore, we inspect the stochastic properties of the errors beyond the first moment in this study.
Note that the null hypothesis can be alternatively stated as: \(\mathcal {H}_{0}^S: F(x) = 1 - F(- x)\) since the forecast errors should be zero-mean stochastic processes. See Footnote 4 for an explanation.
While \(H^{*}\) is unknown, an approximation (of any desired accuracy) can be obtained by Monte Carlo simulation as \(B \rightarrow \infty \).
Theoretically, it would be more policy relevant to assess the distributional properties of the central banks’ forecast errors. Practically, it would be infeasible to compile a comparable dataset of central banks’ forecast errors with the one employed in the study.
Note that quarter-on-quarter percentage changes of economic variables are not considered here since they are available only for GDP but not for CPI.
An alternative way could be to use a Student t distribution with the estimated degrees of freedom which are very likely to be horizon/variable dependent as implied from Fig. 1c.
The Monte Carlo results for different sample sizes are available from the author upon request.
References
Andrews D (1991) Heteroskedasticity and autocorrelation consistent covariance matrix estimation. Econometrica 59:817–858
Aruoba S (2008) Data revisions are not well behaved. J Money Credit Bank 40:319–340
Bai J, Ng S (2005) Tests for skewness, kurtosis, and normality for time series data. J Bus Econ Stat 23:49–60
Balke N, Fomby T (1994) Large shocks, small shocks, and economic fluctuations: outliers in macroeconomic time series. J Appl Econom 9:181–200
Bao Y (2013) On sample skewness and kurtosis. Econom Rev 32:415–448
Berg A, Paparoditis E, Politis DN (2010) A bootstrap test for time series linearity. J Stat Plan Inference 140:3841–3857
Bickel PJ, Bühlmann P (1997) Closure of linear processes. J Theor Probab 10:445–479
Bowman K, Shenton L (1975) Tables of moments of the skewness and kurtosis statistics in non-normal sampling. Union Carbide Nuclear Divison Report
Bühlmann P (1997) Sieve bootstrap for time series. Bernoulli 3:123–148
Chen Y, Lin C (2008) On the robustness of symmetry tests for stock returns. Stud Nonlinear Dyn Econom 12:1–37
Davison A, Hinkley D (1997) Bootstrap methods and their application. Cambridge University Press, Cambridge
Greenspan A (2003) Monetary policy under uncertainty–remarks by chairman alan greenspan at a symposiom of the federal reserve bank of kansas city, vol 19. The Federal Reserve Board, Washington, DC
Haldane AG, Nelson B (2012) Tails of the unexpected. In: Presentation at the conference the credit crisis five years on: unpacking the crisis–University of Edinburgh Business School
Hammond G et al (2012) State of the art of inflation targeting. In: Centre for Central Banking Studies, Bank of England
Harvey DI, Newbold P (2003) The non-normality of some macroeconomic forecast errors. Int J Forecast 19:635–653
Horsewell RL, Looney SW (1993) Diagnostic limitations of skewness coefficients in assessing departures from univariate and multivariate normality. Commun Stat B 22:437–459
Jarque C, Bera A (1980) Efficient tests for normality, homoscedasticity and serial independence of regression residuals. Econ Lett 6:255–259
Jarque C, Bera A (1987) A test for normality of observations and regression residuals. Int Stat Rev 55:163–172
Kim T, White H (2004) On more robust estimation of skewness and kurtosis. Finance Res Lett 1:56–73
Kreiss J-P (1992) Bootstrap procedures for AR\((\infty )\) processes. In: Jöckel K-H, Rothe G, Sendler W (eds) Bootstrapping and related techniques. Springer, Heidelberg, pp 107–113
Lahiri K, Teigland C (1987) On the normality of probability distributions of inflation and gnp forecasts. Int J Forecast 3:269–279
Lee S (1998) On the quantile process based on the autoregressive residuals. J Stat Plan Inference 67:17–28
Lehmann EL, Romano JP (2005) Testing statistical hypotheses. Springer, Berlin
Makridakis S, Winkler RL (1989) Sampling distributions of post-sample forecasting errors. Appl Stat 38:331–342
Müller UK (2014) Hac corrections for strongly autocorrelated time series. J Bus Econ Stat 32:311–322
Paulsen J, Tjøstheim D (1985) On the estimation of residual variance and order in autoregressive time series. J R Stat Soc B 47:216–228
Poskitt DS (2007) Autoregressive approximation in nonstandard situations: the fractionally integrated and non-invertible cases. Ann Inst Stat Math 59:697–725
Poskitt DS (2008) Properties of the sieve bootstrap for fractionally integrated and non-invertible processes. J Time Ser Anal 29:224–250
Premaratne G, Bera A (2005) A test for symmetry with leptokurtic financial data. J Financ Econom 3:169–187
Psaradakis Z, Vávra M (2015) A quantile-based test for symmetry of weakly dependent processes. J Time Ser Anal 36:587–598
Psaradakis Z, Vávra M (2019) Normality tests for dependent data: large-sample and bootstrap approaches. Commun Stat Simul Comput (forthcoming)
Ramberg J, Schmeiser B (1974) An approximate method for generating asymmetric random variables. Commun ACM 17:78–82
Rayner JCW, Best DJ, Mathews KL (1995) Interpreting the skewness coefficient. Commun Stat A 24:593–600
Reifschneider D, Tulip P (2007) Gauging the uncertainty of the economic outlook from historical forecasting errors. Finance Econ Discuss Ser, vol 60
Reifschneider D, Tulip P (2019) Gauging the uncertainty of the economic outlook using historical forecasting errors: the Federal Reserve’s approach. Int J Forecast. (forthcoming)
Sen PK (1968) Asymptotic normality of sample quantiles for m-dependent processes. Ann Math Stat 39:1724–1730
Sharipov OS, Wendler M (2013) Normal limits, nonnormal limits, and the bootstrap for quantiles of dependent data. Stat Probab Lett 83:1028–1035
Shibata R (1980) Asymptotically efficient selection of the order of the model for estimating parameters of a linear process. Ann Stat 8:147–164
Swanepoel J, Van Wyk J (1986) The bootstrap applied to power spectral density function estimation. Biometrika 73:135–141
Tetlow R, Tulip P (2008) Changes in macroeconomic uncertainty. Board of Governors of the Federal Reserve System
Thode H (2002) Testing for normality. Marcel Dekker, NY
Tjøstheim D, Paulsen J (1983) Bias of some commonly-used time series estimates. Biometrika 70:389–399
Tulip P, Wallace S (2012) Estimates of uncertainty around the RBA’s forecasts. Research Discussion Paper, vol 7
Author information
Authors and Affiliations
Corresponding author
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
The author is grateful to Zacharias Psaradakis, Ron Smith, Peter Tulip, and three referees for helpful comments and suggestions in improving an earlier version of the paper.
Appendices
Figures and Tables
See Fig. 2 and Tables 1, 2, 3, and 4.
Simulation study
In this section, we present and discuss the results of a simulation study examining the small-sample properties of the normality and symmetry tests under different patterns of dependence by considering artificial data generated according to the following ARMA models
-
M1:
\(X_{t}=0.5X_{t-1}+\varepsilon _{t}\),
-
M2:
\(X_{t}=0.6X_{t-1}-0.5X_{t-2}+\varepsilon _{t}\),
-
M3:
\(X_{t}=0.6X_{t-1}+0.3\varepsilon _{t-1}+\varepsilon _{t}\).
Here, and throughout this section, \(\{\varepsilon _{t}\}\) are i.i.d. random variables, the common distribution of which is either standard normal (labelled N) or generalized lambda with quantile function \(Q(w)=\lambda _{1}+(1/\lambda _{2})\{w^{\lambda _{3}}-(1-w)^{\lambda _{4}}\}\), \(0<w<1\), standardized to have zero mean and unit variance (see Ramberg and Schmeiser 1974). The parameter values of the generalized lambda distribution used in the experiments are taken from Bai and Ng (2005) and can be found in Table 5, along with the corresponding coefficients of skewness and kurtosis; the distributions N, S1, S2 are symmetric, whereas A1, A2, A3 are asymmetric.
For each design point, 1000 independent realizations of \(\{X_{t}\}\) of length \(n + 100\), with \(n \in \{100,200\}\) (as representative samples for macroeconomic applications), are generated.Footnote 12 The first 100 data points of each realization are then discarded in order to eliminate start-up effects, and the remaining n data points are used to compute the value of the \(\mathcal {T}_N\) and \(\mathcal {T}_S\) test statistics. In the case of bootstrap tests, the order of the autoregressive sieve is determined by minimizing the AIC in the range \(1 \le p < 5\log _{10}(n) \), while the number of bootstrap replications is \(B = 499\). We note that using a larger number of bootstrap replications did not change the results substantially (see Davison and Hinkley 1997, pp. 155–156, for an explanation).
The Monte Carlo rejection frequencies of the test statistics at 5% significance level are reported in Table 6. The null rejection probabilities of the tests are generally insignificantly different from the nominal level across all relevant DGPs. Their rejection frequencies improve with both the sample size and non-normality in the distribution of innovations, although not uniformly (compare the results for A1 and A2). To assess the sensitivity of results with respect to the method used to determine the order of the autoregressive sieve, we consider selecting the latter by minimizing BIC in addition to AIC. The rejection frequencies are reported in Table 7. It is clear that there is little to choose between AIC and BIC, the rejection frequencies not being notably different across the two criteria for any given combination of noise distribution and the sample size n. It is worth noting that results from experiments based on artificial time series confirm the robustness of the properties of the test procedure with respect to the choice of order selection criterion.
Rights and permissions
About this article
Cite this article
Vávra, M. Assessing distributional properties of forecast errors for fan-chart modelling. Empir Econ 59, 2841–2858 (2020). https://doi.org/10.1007/s00181-019-01726-0
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00181-019-01726-0