International Journal of Forecasting最新文献

The aim of this paper is to provide a thinking road-map and a practical guide to researchers and practitioners working on hierarchical forecasting problems. Evaluating the performance of hierarchical forecasts comes with new challenges stemming from both the structure of the hierarchy and the application context. We discuss several relevant dimensions for researchers and analysts: the scale and units of the time series, the issue of intermittency, the forecast horizon, the importance of multiple evaluation windows and the multiple objective decision context. We conclude with a series of practical recommendations.

This paper develops a novel, fully automated forecast averaging scheme which combines LASSO estimation with principal component averaging (PCA). LASSO-PCA (LPCA) explores a pool of predictions based on a single model but calibrated to windows of different sizes. It uses information criteria to select tuning parameters and hence reduces the impact of researchers’ ad hoc decisions. The method is applied to average predictions of hourly day-ahead electricity prices over 650 point forecasts obtained with various lengths of calibration windows. It is evaluated on four European and American markets with an out-of-sample period of almost two and a half years and compared to other semi- and fully automated methods, such as the simple mean, AW/WAW, LASSO, and PCA. The results indicate that LASSO averaging is very efficient in terms of forecast error reduction, whereas PCA is robust to the selection of the specification parameter. LPCA inherits the advantages of both methods and outperforms other approaches in terms of the mean absolute error, remaining insensitive to the choice of a tuning parameter.

Financial crises prediction is an essential topic in finance. Designing an efficient Early Warning System (EWS) can help prevent catastrophic losses resulting from financial crises. We propose different EWSs for predicting potential market instability conditions, where market instability refers to large asset price declines. The EWSs are based on the logit regression and employ Early Warning Indicators (EWIs) based on the realized variance (RV) and/or price-volatility feedback rate. The latter EWI is supposed to describe the ease of the market in absorbing small price perturbations. Our study reveals that, while RV is important in predicting future price losses in a given time series, the EWI employing the price-volatility feedback rate can improve prediction further.

On the one hand, grid integration of solar and wind power often requires just point (as opposed to probabilistic) forecasts at the individual plant level to be submitted to grid operators. On the other hand, solar and wind power forecasting can benefit greatly from dynamical ensemble forecasts from numerical weather prediction (NWP) models. Combining these two facts, this study is concerned with drawing out point forecasts from NWP ensembles. The scoring function for penalizing bad forecasts (or equivalently, rewarding good forecasts), in most scenarios, is specified by grid operators ex ante. The optimal point forecast therefore should be an elicitable functional of the predictive distribution, for which the specified scoring function is strictly consistent. Stated differently, the optimal way to summarize a predictive distribution depends on how the point forecast is to be penalized. Using solar irradiance forecasts issued by the ECMWF’s Ensemble Prediction System, the statistical theory on consistency and elicitability is validated empirically with extensive data. The results show that the optimal point forecasts elicited from ensembles have constantly higher accuracy than the best-guess forecasts, regardless of the choice of scoring function. Surprisingly, however, the correspondence between the two types of goodness of forecasts, namely, quality and value, is neither linear nor monotone, but depends on the penalty triggers and schemes specified by grid operators. In other words, using the optimally elicited forecasts, in many scenarios, would lead to lower economic values.

Auld and Linton (2019) studied the behaviour of the Betfair betting market and the sterling/dollar exchange rate (futures price) during 24 June 2016, the night of the EU referendum. The paper found that both markets appeared to be inefficient, but the currency market was around one hour more inefficient than the betting markets. It has subsequently been discovered that the timestamp used in the betting data was supplied in Greenwich Mean Time as opposed to British Summer Time as assumed by the authors. Updated results suggest that both markets took broadly the same amount of time to discount the public vote information. This calls into doubt the conclusion of a violation of weak market efficiency. Some smaller deviations of the rate at which the markets discount the vote are, however, identified. These were of the order of minutes, suggesting that weak market efficiency did not hold, but to a much smaller degree than first thought.

We develop a flexible modeling framework to produce density nowcasts for US inflation at a trading-day frequency. Our framework (1) combines individual density nowcasts from three classes of parsimonious mixed-frequency models; (2) adopts a novel flexible treatment in the use of the aggregation function; and (3) permits dynamic model averaging via the use of weights that are updated based on learning from past performance. These features provide density nowcasts that can potentially accommodate non-Gaussian properties. We document the competitive properties of the nowcasts generated from our framework using high-frequency real-time data over the period 2000–2015.

This paper introduces a model that can accommodate both the continuous-time-diffusion and discrete-time mixed-GARCH–Jump models by embedding the discrete mixed-GARCH-Jump structure in the continuous volatility process. The key feature of the proposed model is that the corresponding conditional integrated volatility adopts the mixed-GARCH-Jump structure that accounts for the effect of jumps on future volatility. A Griddy–Gibbs sampler approach is proposed to estimate parameters, and volatility forecasting and value-at-risk forecasting based on the peaks-over-threshold method are developed. Simulations are carried out to check the finite sample performance of the proposed methodology, and empirical studies show that, in general, volatility is heavily influenced by the continuous innovations, rather than the extreme reactions. We find that both the simulation and empirical results in most cases support the proposed model.

We propose a novel specification of the Dynamic Conditional Correlation (DCC) model based on an alternative normalization of the pseudo-correlation matrix called Projected DCC (Pro-DCC). Our modification consists in projecting, rather than rescaling, the pseudo-correlation matrix onto the set of correlation matrices in order to obtain a well defined conditional correlation matrix. A simulation study shows that projecting performs better than rescaling when the dimensionality of the correlation matrix is large. An empirical application to the constituents of the S&P 100 shows that the proposed methodology performs favorably to the standard DCC in an out-of-sample asset allocation exercise.

We adapt the multifractal random walk model by Bacry et al. (2001) to realized volatilities (denoted RV-MRW) and take stock of recent theoretical insights on this model in Duchon et al. (2012) to derive forecasts of financial volatility. Moreover, we propose a new extension of the binomial Markov-switching multifractal (BMSM) model by Calvet and Fisher (2001) to the RV framework. We compare the predictive ability of the two against 10 classical and multifractal volatility models. Forecasting performance is evaluated out-of-sample based on the empirical MSE, MAE, and QLIKE criteria as well as using model confidence sets following the methodology of Hansen et al. (2011). Overall, our empirical study for 14 international stock market indices has a clear message: The RV-MRW is the best model throughout under the MAE criterion, i.e., for all indices and forecast horizons between one day and 100 days, with uniform results for forecast evaluations against both realized volatilities and squared returns or during tranquil/turbulent market periods. Based on the evaluation of MSE and QLIKE forecast errors, the RV-MRW, RV-BMSM, and RV-ARFIMA provide the most accurate forecasts during our tranquil sample from 2016–2018, where we can observe a transition from RV-MRW dominating long-term forecasts to RV-BMSM and RV-ARFIMA dominating in the short term. The new RV-BMSM takes the lead in phases of more turbulent market dynamics (sample 2010–2012), when it appears throughout in the 90% model confidence set at horizons $\le 10$ days and for 13 out of 14 indices at 20 days. These results are very promising if we consider that this is the first empirical application of the RV-MRW and RV-BMSM. Moreover, whereas RV-ARFIMA forecasts are often a time-consuming task, the RV-MRW stands out due to its fast execution and straightforward implementation.