GEFCom2012: Electric load forecasting and backcasting with semi-parametric models

We sum up the methodology of the team tololo for the Global Energy Forecasting Competition 2012: Load Forecasting. Our strategy consisted of a temporal multi-scale model that combines three components. The first component was a long term trend estimated by means of non-parametric smoothing. The second was a medium term component describing the sensitivity of the electricity demand to the temperature at each time step. We use a generalized additive model to fit this component, using calendar information as well. Finally, a short term component models local behaviours. As the factors that drive this component are unknown, we use a random forest model to estimate it.     

A new structural break model,with an application to Canadian inflation forecasting

This paper develops an efficient approach to modelling and forecasting time series data with an unknown number of change-points. Using a conjugate prior and conditioning on time-invariant parameters, the predictive density and the posterior distribution of the change-points have closed forms. Furthermore, the conjugate prior is modeled as hierarchical in order to exploit the information across regimes. This framework allows breaks in the variance, the regression coefficients, or both. The regime duration can be modelled as a Poisson distribution. A new, efficient Markov chain Monte Carlo sampler draws the parameters from the posterior distribution as one block. An application to a Canadian inflation series shows the gains in forecasting precision that our model provides.     

A data-driven approach to forecasting ground-level ozone concentration

The ability to forecast the concentration of air pollutants in an urban region is crucial for decision-makers wishing to reduce the impact of pollution on public health through active measures (e.g. temporary traffic closures). In this study, we present a machine learning approach applied to forecasts of the day-ahead maximum value of ozone concentration for several geographical locations in southern Switzerland. Due to the low density of measurement stations and to the complex orography of the use-case terrain, we adopted feature selection methods instead of explicitly restricting relevant features to a neighborhood of the prediction sites, as common in spatio-temporal forecasting methods. We then used Shapley values to assess the explainability of the learned models in terms of feature importance and feature interactions in relation to ozone predictions. Our analysis suggests that the trained models effectively learned explanatory cross-dependencies among atmospheric variables. Finally, we show how weighting observations helps to increase the accuracy of the forecasts for specific ranges of ozone’s daily peak values.     

Forecasting of density functions with an application to cross-sectional and intraday returns

This paper is concerned with the forecasting of probability density functions. Density functions are nonnegative and have a constrained integral, and thus do not constitute a vector space. The implementation of established functional time series forecasting methods for such nonlinear data is therefore problematic. Two new methods are developed and compared to two existing methods. The comparison is based on the densities derived from cross-sectional and intraday returns. For such data, one of our new approaches is shown to dominate the existing methods, while the other is comparable to one of the existing approaches.     

Forecasting the NN5 time series with hybrid models

We propose a simple way of predicting time series with recurring seasonal periods. Missing values of the time series are estimated and interpolated in a preprocessing step. We combine several forecasting methods by taking the weighted mean of forecasts that were generated with time-domain models which were validated on left-out parts of the time series. The hybrid model is a combination of a neural network ensemble, an ensemble of nearest trajectory models and a model for the 7-day cycle. We apply this approach to the NN5 time series competition data set.     

Solving DSGE models with a nonlinear moving average

We propose a nonlinear infinite moving average as an alternative to the standard state space policy function for solving nonlinear DSGE models. Perturbation of the nonlinear moving average policy function provides a direct mapping from a history of innovations to endogenous variables, decomposes the contributions from individual orders of uncertainty and nonlinearity, and enables familiar impulse response analysis in nonlinear settings. When the linear approximation is saddle stable and free of unit roots, higher order terms are likewise saddle stable and first order corrections for uncertainty are zero. We derive the third order approximation explicitly, examine the accuracy of the method using Euler equation tests, and compare with state space approximations.     

Pair trading based on quantile forecasting of smooth transition GARCH models

Pair trading is a statistical arbitrage strategy used on similar assets with dissimilar valuations. We utilize smooth transition heteroskedastic models with a second-order logistic function to generate trading entry and exit signals and suggest two pair trading strategies: the first uses the upper and lower threshold values in the proposed model as trading entry and exit signals, while the second strategy instead takes one-step-ahead quantile forecasts obtained from the same model. We employ Bayesian Markov chain Monte Carlo sampling methods for updating the estimates and quantile forecasts. As an illustration, we conduct a simulation study and empirical analysis of the daily stock returns of 36 stocks from U.S. stock markets. We use the minimum square distance method to select ten stock pairs, choose additional five pairs consisting of two companies in the same industrial sector, and then finally consider pair trading profits for two out-of-sample periods in 2014 within a six-month time frame as well as for the entire year. The proposed strategies yield average annualized returns of at least 35.5% without a transaction cost and at least 18.4% with a transaction cost.     

Multivariate semi-nonparametric distributions with dynamic conditional correlations

This paper generalizes the Dynamic Conditional Correlation (DCC) model of Engle (2002), incorporating a flexible non-Gaussian distribution based on Gram-Charlier expansions. The resulting semi-nonparametric-DCC (SNP-DCC) model allows estimation in two stages and deals with the negativity problem which is inherent in truncated SNP densities. We test the performance of a SNP-DCC model with respect to the (Gaussian)-DCC through an empirical application of density forecasting for portfolio returns. Our results show that the proposed multivariate model provides a better in-sample fit and forecast of the portfolio returns distribution, and thus is useful for financial risk forecasting and evaluation.     

Forecasting correlated time series with exponential smoothing models

This paper presents the Bayesian analysis of a general multivariate exponential smoothing model that allows us to forecast time series jointly, subject to correlated random disturbances. The general multivariate model, which can be formulated as a seemingly unrelated regression model, includes the previously studied homogeneous multivariate Holt-Winters’ model as a special case when all of the univariate series share a common structure. MCMC simulation techniques are required in order to approach the non-analytically tractable posterior distribution of the model parameters. The predictive distribution is then estimated using Monte Carlo integration. A Bayesian model selection criterion is introduced into the forecasting scheme for selecting the most adequate multivariate model for describing the behaviour of the time series under study. The forecasting performance of this procedure is tested using some real examples.     

Combination of long term and short term forecasts, with application to tourism demand forecasting

Forecast combination is a well-established and well-tested approach for improving the forecasting accuracy. One beneficial strategy is to use constituent forecasts that have diverse information. In this paper we consider the idea of diversity being accomplished by using different time aggregations. For example, we could create a yearly time series from a monthly time series and produce forecasts for both, then combine the forecasts. These forecasts would each be tracking the dynamics of different time scales, and would therefore add diverse types of information. A comparison of several forecast combination methods, performed in the context of this setup, shows that this is indeed a beneficial strategy and generally provides a forecasting performance that is better than the performances of the individual forecasts that are combined.As a case study, we consider the problem of forecasting monthly tourism numbers for inbound tourism to Egypt. Specifically, we consider 33 individual source countries, as well as the aggregate. The novel combination strategy also produces a generally improved forecasting accuracy.     

Stochastic volatility models with ARMA innovations: An application to G7 inflation forecasts

We introduce a new class of stochastic volatility models with autoregressive moving average (ARMA) innovations. The conditional mean process has a flexible form that can accommodate both a state space representation and a conventional dynamic regression. The ARMA component introduces serial dependence, which results in standard Kalman filter techniques not being directly applicable. To overcome this hurdle, we develop an efficient posterior simulator that builds on recently developed precision-based algorithms. We assess the usefulness of these new models in an inflation forecasting exercise across all G7 economies. We find that the new models generally provide competitive point and density forecasts compared to standard benchmarks, and are especially useful for Canada, France, Italy, and the U.S.     

Open economy forecasting with a DSGE-VAR: Head to head with the RBNZ published forecasts

We construct a DSGE-VAR model for competing head to head with the long history of published forecasts of the Reserve Bank of New Zealand. We also construct a Bayesian VAR model with a Minnesota prior for forecast comparison. The DSGE-VAR model combines a structural DSGE model with a statistical VAR model based on the in-sample fit over the majority of New Zealand’s inflation-targeting period. We evaluate the real-time out-of-sample forecasting performance of the DSGE-VAR model, and show that the forecasts from the DSGE-VAR are competitive with the Reserve Bank of New Zealand’s published, judgmentally-adjusted forecasts. The Bayesian VAR model with a Minnesota prior also provides a competitive forecasting performance, and generally, with a few exceptions, out-performs both the DSGE-VAR and the Reserve Bank’s own forecasts.     

A nonlinear long memory model, with an application to US unemployment

Two important empirical features of US unemployment are that shocks to the series seem rather persistent and that it seems to rise faster during recessions than that it falls during expansions. To jointly capture these features of long memory and nonlinearity, we put forward a new time series model and evaluate its empirical performance. We find that the model describes the data rather well and that it outperforms related competitive models on various measures of fit.     

Local influence analysis for Poisson autoregression with an application to stock transaction data

In statistical diagnostics and sensitivity analysis, the local influence method plays an important role and has certain advantages over other methods in several situations. In this paper, we use this method to study time series of count data when employing a Poisson autoregressive model. We consider case‐weights, scale, data, and additive perturbation schemes to obtain their corresponding vectors and matrices of derivatives for the measures of slope and normal curvatures. Based on the curvature diagnostics, we take a stepwise local influence approach to deal with data with possible masking effects. Finally, our established results are illustrated to be effective by analyzing a stock transactions data set.