Evaluating quantile forecasts in the M5 uncertainty competition

Probabilistic forecasts are necessary for robust decisions in the face of uncertainty. The M5 Uncertainty competition required participating teams to forecast nine quantiles for unit sales of various products at various aggregation levels and for different time horizons. This paper evaluates the forecasting performance of the quantile forecasts at different aggregation levels and at different quantile levels. We contrast this with some theoretical predictions, and discuss potential implications and promising future research directions for the practice of probabilistic forecasting.     

Machine learning methods for GEFCom2017 probabilistic load forecasting

This paper describes the preprocessing and forecasting methods used by team Orbuculum during the qualifying match of the Global Energy Forecasting Competition 2017 (GEFCom2017). Tree-based algorithms (gradient boosting and quantile random forest) and neural networks made up an ensemble. The ensemble prediction quantiles were obtained by a simple averaging of the ensemble members’ prediction quantiles. The result shows a robust performance according to the pinball loss metric, with the ensemble model achieving third place in the qualifying match of the competition.     

Partial cross-quantilogram networks: Measuring quantile connectedness of financial institutions

We propose partial cross-quantilogram networks for measuring the connectedness of 30 China’s financial institutions at different quantiles. We find that networks at the extreme quantiles are more closely connected than those at the median quantile. The network density and centrality show that the systemically important financial institutions vary across different quantiles. We observe an asymmetric effect in quantile connectedness during the period of “2015–16 Chinese stock market turbulence;” that is, the network connectedness at the lower quantile (i.e., 0.05 quantile) is higher than that at the upper and median quantiles (i.e., 0.95 and 0.50 quantiles). By analyzing the similarity of networks across quantiles, we find that the similarity index is relatively high in the crisis period. Our study provides useful information on connectedness of financial institutions for regulators and investors.     

Quasi-maximum likelihood estimation for conditional quantiles

In this paper, we construct a new class of estimators for conditional quantiles in possibly misspecified nonlinear models with time series data. Proposed estimators belong to the family of quasi-maximum likelihood estimators (QMLEs) and are based on a new family of densities which we call ‘tick-exponential’. A well-known member of the tick-exponential family is the asymmetric Laplace density, and the corresponding QMLE reduces to the Koenker and Bassett's (Econometrica 46 (1978) 33) nonlinear quantile regression estimator. We derive primitive conditions under which the tick-exponential QMLEs are consistent and asymptotically normally distributed with an asymptotic covariance matrix that accounts for possible conditional quantile model misspecification and which can be consistently estimated by using the tick-exponential scores and Hessian matrix. Despite its non-differentiability, the tick-exponential quasi-likelihood is easy to maximize by using a ‘minimax’ representation not seen in the earlier work on conditional quantile estimation.     

Quantile models and estimators for data analysis

Quantile regression is used to estimate the cross sectional relationship between high school characteristics and student achievement as measured by ACT scores. The importance of school characteristics on student achievement has been traditionally framed in terms of the effect on the expected value. With quantile regression the impact of school characteristics is allowed to be different at the mean and quantiles of the conditional distribution. Like robust estimation, the quantile approach detects relationships missed by traditional data analysis. Robust estimates detect the influence of the bulk of the data, whereas quantile estimates detect the influence of co-variates on alternate parts of the conditional distribution. Since our design consists of multiple responses (individual student ACT scores) at fixed explanatory variables (school characteristics) the quantile model can be estimated by the usual regression quantiles, but additionally by a regression on the empirical quantile at each school. This is similar to least squares where the estimate based on the entire data is identical to weighted least squares on the school averages. Unlike least squares however, the regression through the quantiles produces a different estimate than the regression quantiles.     

Realized extreme quantile: A joint model for conditional quantiles and measures of volatility with EVT refinements

We propose a new framework exploiting realized measures of volatility to estimate and forecast extreme quantiles. Our realized extreme quantile (REQ) combines quantile regression with extreme value theory and uses a measurement equation that relates the realized measure to the latent conditional quantile. Model estimation is performed by quasi maximum likelihood, and a simulation experiment validates this estimator in finite samples. An extensive empirical analysis shows that high‐frequency measures are particularly informative of the dynamic quantiles. Finally, an out‐of‐sample forecast analysis of quantile‐based risk measures confirms the merit of the REQ.     

Nonparametric expected shortfall forecasting incorporating weighted quantiles

A new semi-parametric expected shortfall (ES) estimation and forecasting framework is proposed. The proposed approach is based on a two-step estimation procedure. The first step involves the estimation of value at risk (VaR) at different quantile levels through a set of quantile time series regressions. Then, the ES is computed as a weighted average of the estimated quantiles. The quantile weighting structure is parsimoniously parameterized by means of a beta weight function whose coefficients are optimized by minimizing a joint VaR and ES loss function of the Fissler–Ziegel class. The properties of the proposed approach are first evaluated with an extensive simulation study using two data generating processes. Two forecasting studies with different out-of-sample sizes are then conducted, one of which focuses on the 2008 Global Financial Crisis period. The proposed models are applied to seven stock market indices, and their forecasting performances are compared to those of a range of parametric, non-parametric, and semi-parametric models, including GARCH, conditional autoregressive expectile (CARE), joint VaR and ES quantile regression models, and a simple average of quantiles. The results of the forecasting experiments provide clear evidence in support of the proposed models.     

Testing for structural change in regression quantiles

Most studies in the structural change literature focus solely on the conditional mean, while under various circumstances, structural change in the conditional distribution or in conditional quantiles is of key importance. This paper proposes several tests for structural change in regression quantiles. Two types of statistics are considered, namely, a fluctuation type statistic based on the subgradient and a Wald type statistic, based on comparing parameter estimates obtained from different subsamples. The former requires estimating the model under the null hypothesis, and the latter involves estimation under the alternative hypothesis. The tests proposed can be used to test for structural change occurring in a pre-specified quantile, or across quantiles, which can be viewed as testing for change in the conditional distribution with a linear specification of the conditional quantile function. Both single and multiple structural changes are considered. We derive the limiting distributions under the null hypothesis, and show they are nuisance parameter free and can be easily simulated. A simulation study is conducted to assess the size and power in finite samples.     

Quantiles, expectiles and splines

A time-varying quantile can be fitted by formulating a time series model for the corresponding population quantile and iteratively applying a suitably modified state space signal extraction algorithm. It is shown that such quantiles satisfy the defining property of fixed quantiles in having the appropriate number of observations above and below. Like quantiles, time-varying expectiles can be estimated by a state space signal extraction algorithm and they satisfy properties that generalize the moment conditions associated with fixed expectiles. Because the state space form can handle irregularly spaced observations, the proposed algorithms can be adapted to provide a viable means of computing spline-based non-parametric quantile and expectile regressions.     

基于Stacking模型融合的用户购买行为预测研究

在大数据时代背景下,如何利用大量的销售数据精准预测顾客未来需求,成为企业制定客户管理和库存管理决策的一个重要问题。目前关于用户购买行为预测的研究中很少能够预测用户具体的购买时间。基于已有的销售数据,提出了基于机器学习和Stacking集成的综合预测模型预测用户的购买行为,即未来是否购买及其购买时间。将模型应用在一家大型连锁零售企业的需求预测中,并对方法的有效性进行评估。结果表明,基于Stacking集成的融合模型对预测用户未来是否购买具有最佳性能,准确率达85%,AUC值达到0.928;LightGBM集成算法在预测用户购买时间时具有最优性能,相比于融合模型提升了5.5%的预测性能;融合模型+LightGBM算法的组合相比于均使用融合模型提升了9.4%的预测性能。     

Forecasting from others’ experience: Bayesian estimation of the generalized Bass model

We propose a Bayesian estimation procedure for the generalized Bass model that is used in product diffusion models. Our method forecasts product sales early based on previous similar markets; that is, we obtain pre-launch forecasts by analogy. We compare our forecasting proposal to traditional estimation approaches, and alternative new product diffusion specifications. We perform several simulation exercises, and use our method to forecast the sales of room air conditioners, BlackBerry handheld devices, and compressed natural gas. The results show that our Bayesian proposal provides better predictive performances than competing alternatives when little or no historical data are available, which is when sales projections are the most useful.     

Actual and counterfactual growth incidence and delta Lorenz curves: Estimation and inference

Different economic growth episodes display very different distributional characteristics, both across countries and over time. Growth is sometimes accompanied by rising and sometimes by falling inequality. Applied economists have come to rely on the growth incidence curve, which gives the quantile‐specific rate of income growth over a certain period, to describe these differences. This paper introduces a mean‐independent analogue, the delta Lorenz curve, which gives the cumulative change in income share up to each quantile. We also develop estimation and inference procedures for both functions of quantiles. We establish the limiting null distribution of the test statistics of interest for those functions, and propose resampling methods to implement inference in practice. The proposed methods are used to compare the growth processes in the USA and Brazil during 1995–2007. Although growth in the average real wages was disappointing in both countries, the distribution of that growth was markedly different. In the USA, wage growth was mediocre for the bottom 80% of the sample, but much more rapid for the top 20%. In Brazil, conversely, wage growth was rapid below the median, and negative at the top. Wage shares fell in the USA up to the 83rd percentile, and rose in Brazil up to the 65th percentile.     

Quantile cointegrating regression

Quantile regression has important applications in risk management, portfolio optimization, and asset pricing. The current paper studies estimation, inference and financial applications of quantile regression with cointegrated time series. In addition, a new cointegration model with quantile-varying coefficients is proposed. In the proposed model, the value of cointegrating coefficients may be affected by the shocks and thus may vary over the innovation quantile. The proposed model may be viewed as a stochastic cointegration model which includes the conventional cointegration model as a special case. It also provides a useful complement to cointegration models with (G)ARCH effects. Asymptotic properties of the proposed model and limiting distribution of the cointegrating regression quantiles are derived. In the presence of endogenous regressors, fully-modified quantile regression estimators and augmented quantile cointegrating regression are proposed to remove the second order bias and nuisance parameters. Regression Wald tests are constructed based on the fully modified quantile regression estimators. An empirical application to stock index data highlights the potential of the proposed method.     

Hierarchical forecasting with a top-down alignment of independent-level forecasts

Hierarchical forecasting with intermittent time series is a challenge in both research and empirical studies. Extensive research focuses on improving the accuracy of each hierarchy, especially the intermittent time series at bottom levels. Then, hierarchical reconciliation can be used to improve the overall performance further. In this paper, we present a hierarchical-forecasting-with-alignment approach that treats the bottom-level forecasts as mutable to ensure higher forecasting accuracy on the upper levels of the hierarchy. We employ a pure deep learning forecasting approach, N-BEATS, for continuous time series at the top levels, and a widely used tree-based algorithm, LightGBM, for intermittent time series at the bottom level. The hierarchical-forecasting-with-alignment approach is a simple yet effective variant of the bottom-up method, accounting for biases that are difficult to observe at the bottom level. It allows suboptimal forecasts at the lower level to retain a higher overall performance. The approach in this empirical study was developed by the first author during the M5 Accuracy competition, ranking second place. The method is also business orientated and can be used to facilitate strategic business planning.     

The quantilogram: With an application to evaluating directional predictability

We propose a new diagnostic tool for time series called the quantilogram. The tool can be used formally and we provide the inference tools to do this under general conditions, and it can also be used as a simple graphical device. We apply our method to measure directional predictability and to test the hypothesis that a given time series has no directional predictability. The test is based on comparing the correlogram of quantile hits to a pointwise confidence interval or on comparing the cumulated squared autocorrelations with the corresponding critical value. We provide the distribution theory needed to conduct inference, propose some model free upper bound critical values, and apply our methods to S&P500 stock index return data. The empirical results suggest some directional predictability in returns. The evidence is strongest in mid range quantiles like 5–10% and for daily data. The evidence for predictability at the median is of comparable strength to the evidence around the mean, and is strongest at the daily frequency.     

分位数局部调整模型及应用

本文基于分位数的回归理论与方法,提出了一个新的经济计量模型:分位数局部调整模型,并给出了其数学表示、参数估计与预测方法等一整套建模技术。分位数局部调整模型能够细致地给出响应变量在各个分位点上的条件分位数,便于揭示响应变量位置、散布与形状等动态调整过程的全景信息,从而得到比均值局部调整模型更为深刻的结果。最后,将分位数局部调整模型应用于中国货币需求分析,结果显示,在货币需求的不同阶段,不仅调整速度不同,调整方式也呈现出非对称性;M1存在货币失踪之谜现象,而M2却在条件密度第一个最优区域实现了供求均衡;最优货币需求条件密度曲线较为分散,这为央行制定货币政策预留了足够的空间。     

Markov switching quantile autoregression

This paper considers the location‐scale quantile autoregression in which the location and scale parameters are subject to regime shifts. The regime changes in lower and upper tails are determined by the outcome of a latent, discrete‐state Markov process. The new method provides direct inference and estimate for different parts of a non‐stationary time series distribution. Bayesian inference for switching regimes within a quantile, via a three‐parameter asymmetric Laplace distribution, is adapted and designed for parameter estimation. Using the Bayesian output, the marginal likelihood is readily available for testing the presence and the number of regimes. The simulation study shows that the predictability of regimes and conditional quantiles by using asymmetric Laplace distribution as the likelihood is fairly comparable with the true model distributions. However, ignoring that autoregressive coefficients might be quantile dependent leads to substantial bias in both regime inference and quantile prediction. The potential of this new approach is illustrated in the empirical applications to the US inflation and real exchange rates for asymmetric dynamics and the S&P 500 index returns of different frequencies for financial market risk assessment.     

Nonparametric estimation and inference on conditional quantile processes

This paper presents estimation methods and asymptotic theory for the analysis of a nonparametrically specified conditional quantile process. Two estimators based on local linear regressions are proposed. The first estimator applies simple inequality constraints while the second uses rearrangement to maintain quantile monotonicity. The bandwidth parameter is allowed to vary across quantiles to adapt to data sparsity. For inference, the paper first establishes a uniform Bahadur representation and then shows that the two estimators converge weakly to the same limiting Gaussian process. As an empirical illustration, the paper considers a dataset from Project STAR and delivers two new findings.     

Estimation of copula-based semiparametric time series models

This paper studies the estimation of a class of copula-based semiparametric stationary Markov models. These models are characterized by nonparametric marginal distributions and parametric copula functions, while the copulas capture all the scale-free temporal dependence of the processes. Simple estimators of the marginal distribution and the copula parameter are provided, and their asymptotic properties are established under easily verifiable conditions. These results are used to obtain root-n consistent and asymptotically normal estimators of important features of the transition distribution such as the (nonlinear) conditional moments and conditional quantiles. The semiparametric conditional quantile estimators are automatically monotonic across quantiles, which is attractive for portfolio conditional value-at-risk calculations.