Temporal Fusion Transformers for interpretable multi-horizon time series forecasting

Multi-horizon forecasting often contains a complex mix of inputs – including static (i.e. time-invariant) covariates, known future inputs, and other exogenous time series that are only observed in the past – without any prior information on how they interact with the target. Several deep learning methods have been proposed, but they are typically ‘black-box’ models that do not shed light on how they use the full range of inputs present in practical scenarios. In this paper, we introduce the Temporal Fusion Transformer (TFT) – a novel attention-based architecture that combines high-performance multi-horizon forecasting with interpretable insights into temporal dynamics. To learn temporal relationships at different scales, TFT uses recurrent layers for local processing and interpretable self-attention layers for long-term dependencies. TFT utilizes specialized components to select relevant features and a series of gating layers to suppress unnecessary components, enabling high performance in a wide range of scenarios. On a variety of real-world datasets, we demonstrate significant performance improvements over existing benchmarks, and highlight three practical interpretability use cases of TFT.     

Evaluating strange forecasts: The curious case of football match scorelines

This study analyses point forecasts of exact scoreline outcomes for football matches in the English Premier League. These forecasts were made for distinct competitions and originally judged differently. We compare these with implied probability forecasts using bookmaker odds and a crowd of tipsters, as well as point and probability forecasts generated from a statistical model. From evaluating these sources and types of forecast, using various methods, we argue that regression encompassing is the most appropriate way to compare point and probability forecasts, and find that both these types of forecasts for football match scorelines generally add information to one another.     

Forecasting spot prices of agricultural commodities in India: Application of deep-learning models

Food price fluctuations can impact both producers and consumers. Forecasting the prices of the agricultural commodities is of prime concern not only to the government but also to farmers and agribusiness firms. In developing countries like India, management of food security needs competent and efficient forecasting of food prices. With the availability of data, recent innovation in deep-learning models provides a feasible solution to accurately forecast the prices. In this study, we examine the superiority of these models using the daily spot prices of five major commodities traded on the National Commodity and Derivatives Exchange: cotton seed, castor seed, rape mustard seed, soybean seed, and guar seed. The results were obtained from the application of the traditional univariate autoregressive integrated moving average model and deep-learning techniques like the time-delay neural network (TDNN) and long short-term memory (LSTM) network. The empirical results indicate that the LSTM model is indeed suitable for the financial domain and captures the directional movement of the spot price changes with high accuracy compared with the TDNN and other linear models. Accuracy of the performance of these models has been compared using out-of-sample performance measure. The overall objective of this paper is to demonstrate the utility of spot price forecasting for farmers and traders in offering them the best predictions of the price movements. Our results provide a possibility of developing pricing models that can help in fairly regulating agricultural commodity prices.     

Modelling and forecasting national introduction times for successive generations of mobile telephony

An accurate prediction of the timing of a country's introduction of a new generation of mobile telephony benefits numerous agents including suppliers of network and consumer equipment, regulators, and network planners. We consider the estimation and prediction of the time interval between the international introduction of a generation of mobile telephony and its introduction into a specific country when a decision maker judges the introduction of a newer technology a worthwhile investment. Using literature-based socio-economic and geographical variables, we examine how well variation in international introduction times of four generations of mobile telephony in 172 countries can be explained and forecast. We model and forecast introduction times at two levels of granularity: we use Cox's proportional hazards model for the introduction time; we partition countries into introduction time-based segments and model segment membership using multinomial logistic regression. Our modelling of each generation considers three subsets of explanatory variables: All variables, socio-economic Covariates only, Regional dummies only. Over successive generations, the Covariates only models reveal the changing relevance of each socio-economic covariate. Model-based forecasting of the introduction time of the next generation is performed under three hypotheses making different uses of the information available at the time the relevant generation is launched internationally. However, changing socio-economic environments coupled with changing models impair forecasting accuracy, the lower accuracy of modelled introduction times is concentrated in 20% of countries. We speculate about the nature of the unobserved factors affecting these countries' decision processes.     

Novel deep learning approach for forecasting daily hotel demand with agglomeration effect

Increasing competition and adoption of revenue management practices in the hotel industry fuel the need for accurate forecasting to maximize profits and optimize operations. Considering the limitations of relevant research, this study focuses on the daily hotel demand with consideration of agglomeration effect, and proposes a novel deep learning-based model, namely, Deep Learning Model with Spatial and Temporal correlations. This model contributes to relevant research by introducing the agglomeration effect and integrating the attention mechanism and Bayesian optimization algorithm. Historical daily demand data of 210 hotels in Xiamen, China are used to verify the model performance. Results show that the proposed model is significantly better than the benchmarks. This study can help hotel managers improve revenue management through better matching potential demand to available capacity.     

Forecasting volatility of crude oil futures using a GARCH–RNN hybrid approach

Volatility is an important element for various financial instruments owing to its ability to measure the risk and reward value of a given financial asset. Owing to its importance, forecasting volatility has become a critical task in financial forecasting. In this paper, we propose a suite of hybrid models for forecasting volatility of crude oil under different forecasting horizons. Specifically, we combine the parameters of generalized autoregressive conditional heteroscedasticity (GARCH) and Glosten–Jagannathan–Runkle (GJR)-GARCH with long short-term memory (LSTM) to create three new forecasting models named GARCH–LSTM, GJR-LSTM, and GARCH-GJRGARCH LSTM in order to forecast crude oil volatility of West Texas Intermediate on different forecasting horizons and compare their performance with the classical volatility forecasting models. Specifically, we compare the performances against existing methodologies of forecasting volatility such as GARCH and found that the proposed hybrid models improve upon the forecasting accuracy of Crude Oil: West Texas Intermediate under various forecasting horizons and perform better than GARCH and GJR-GARCH, with GG–LSTM performing the best of the three proposed models at 7-, 14-, and 21-day-ahead forecasts in terms of heteroscedasticity-adjusted mean square error and heteroscedasticity-adjusted mean absolute error, with significance testing conducted through the model confidence set showing that GG–LSTM is a strong contender for forecasting crude oil volatility under different forecasting regimes and rolling-window schemes. The contribution of the paper is that it enhances the forecasting ability of crude oil futures volatility, which is essential for trading, hedging, and purposes of arbitrage, and that the proposed model dwells upon existing literature and enhances the forecasting accuracy of crude oil volatility by fusing a neural network model with multiple econometric models.     

The M5 uncertainty competition: Results,findings and conclusions

This paper describes the M5 “Uncertainty” competition, the second of two parallel challenges of the latest M competition, aiming to advance the theory and practice of forecasting. The particular objective of the M5 “Uncertainty” competition was to accurately forecast the uncertainty distributions of the realized values of 42,840 time series that represent the hierarchical unit sales of the largest retail company in the world by revenue, Walmart. To do so, the competition required the prediction of nine different quantiles (0.005, 0.025, 0.165, 0.250, 0.500, 0.750, 0.835, 0.975, and 0.995), that can sufficiently describe the complete distributions of future sales. The paper provides details on the implementation and execution of the M5 “Uncertainty” competition, presents its results and the top-performing methods, and summarizes its major findings and conclusions. Finally, it discusses the implications of its findings and suggests directions for future research.     

Do commercial real estate prices have predictive content for GDP?

Using a uniquely compiled database concerning rental prices of commercial real estates, which are property of the largest broker in the Netherlands, we examine whether these prices have predictive value for quarterly economic growth. In contrast to related studies, we document that the mean price contains no relevant information, whereas other properties of the price distributions have. We show that these distributions can be described by mixtures of two distributions, reflecting low-end and high-end price segments. Our main findings are that higher economic growth is predictable from more new buildings being rented, more variation in the price levels and a larger size of the low-price segment, while lower economic growth emerges when the differences in prices between high-end and low-end segments increase and when the average price level in the low-price segment increases.     

Why is it so difficult to outperform the random walk in exchange rate forecasting?

A simulation exercise is used to demonstrate the difficulty to outperform the random walk in exchange rate forecasting if forecasting accuracy is judged by the Root Mean Square Error (RMSE) or similar criteria that depend on the magnitude of the forecasting error. It is shown that, as the exchange rate volatility rises, the RMSE of the model rises faster than that of the random walk. While the literature considers this finding to be a puzzle that casts a big shadow of doubt on the soundness of international monetary economics, the results show that failure to outperform the random walk, in both in-sample and out-of-sample forecasting, should be the rule rather than the exception. However, the results do not imply that the random walk is unbeatable, because it can be easily outperformed if forecasting accuracy is judged according to criteria such as direction accuracy and profitability.     

Location,location, location: currency effects and return predictability?

Most international financial market studies that compare across countries utilize the US dollar as the common numeraire. We explore the little studied question of the appropriate choice for the base currency and ask if currency choice can affect the final conclusion of whether predictability exists. We provide empirical results for stock return predictability that demonstrate the importance of the numeraire. For example, the existence (absence) of predictability for a US investor does not necessarily imply the existence (absence) of predictability for other foreign investors.