|
on Forecasting |
| By: | Pesaran, M. H.; Smith, R. P. |
| Abstract: | Forecasts play a central role in decision making under uncertainty. After a brief review of the general issues, this paper considers ways of using high-dimensional data in forecasting. We consider selecting variables from a known active set, known knowns, using Lasso and OCMT, and approximating unobserved latent factors, known unknowns, by various means. This combines both sparse and dense approaches. We demonstrate the various issues involved in variable selection in a high-dimensional setting with an application to forecasting UK inflation at different horizons over the period 2020q1-2023q1. This application shows both the power of parsimonious models and the importance of allowing for global variables. |
| Keywords: | Forecasting, high-dimensional data, Lasso, OCMT, latent factors, principal components |
| JEL: | C53 C55 E37 E52 |
| Date: | 2024–02–06 |
| URL: | http://d.repec.org/n?u=RePEc:cam:camdae:2406&r=for |
| By: | Xinyi Wang; Lang Tong |
| Abstract: | This paper presents a novel generative probabilistic forecasting approach derived from the Wiener-Kallianpur innovation representation of nonparametric time series. Under the paradigm of generative artificial intelligence, the proposed forecasting architecture includes an autoencoder that transforms nonparametric multivariate random processes into canonical innovation sequences, from which future time series samples are generated according to their probability distributions conditioned on past samples. A novel deep-learning algorithm is proposed that constrains the latent process to be an independent and identically distributed sequence with matching autoencoder input-output conditional probability distributions. Asymptotic optimality and structural convergence properties of the proposed generative forecasting approach are established. Three applications involving highly dynamic and volatile time series in real-time market operations are considered: (i) locational marginal price forecasting for merchant storage participants, {(ii) interregional price spread forecasting for interchange markets, } and (iii) area control error forecasting for frequency regulations. Numerical studies based on market data from multiple independent system operators demonstrate superior performance against leading traditional and machine learning-based forecasting techniques under both probabilistic and point forecast metrics. |
| Date: | 2024–03 |
| URL: | http://d.repec.org/n?u=RePEc:arx:papers:2403.05743&r=for |
| By: | Salih Zeki Atilgan; Tarik Aydogdu; Mehmet Selman Colak; Muhammed Hasan Yilmaz |
| Abstract: | In this paper, we propose the use of regularization and shrinkage methods to address the variable selection problem in predicting credit growth. Using data from the 10 largest Turkish banks and a broader set of macro-financial predictors for the period 2012-2023, we find that the models generated by the Least Absolute Shrinkage and Selection Operator (LASSO) method have superior predictive power (lower level of forecast errors) for bank-level total credit growth compared to alternative factor-augmented models through recursive out-of-sample forecasting exercises. Our baseline findings remain intact against alternative choices of the tuning parameter and LASSO specifications. In addition to the dynamics of the total credit growth, the improvement in prediction accuracy is evident for commercial credit growth at all horizons, while the effect is limited to short-term horizons for consumer credit growth. Furthermore, additional robustness checks show that the baseline results do not vary considerably against different sample coverage and benchmark models. In the subsequent analyses, we utilize the LASSO method to synthesize the “residual credit” indicator as a proxy for excessive credit movements deviating from the level implied by macro-financial dynamics. In the scope of a case study, using this indicator as an input for local projection estimates, we show that recent inflationary pressures have resulted in excessive lending activity, which is not fully explained by macro-financial dynamics, in the period 2020-2023. |
| Keywords: | Credit growth, Forecasting, LASSO, Residual credit, Local projection |
| JEL: | G21 C53 C55 |
| Date: | 2024 |
| URL: | http://d.repec.org/n?u=RePEc:tcb:wpaper:2402&r=for |
| By: | Ashby, M.; Harvey, A.; Kattuman, P.; Thamotheram, C. |
| Abstract: | This paper documents the Time Series Growth Curves (tsgc) package for R, which is designed for forecasting epidemics, including the detection of new waves and turning points. The package implements time series growth curve methods founded on a dynamic Gompertz model and can be estimated using techniques based on state space models and the Kalman filter. The model is suitable for predicting future values of any variable which, when cumulated, is subject to some unknown saturation level. In the context of epidemics, the model can adjust to changes in social behavior and policy. It is also relevant for many other domains, such as the diffusion of new products. The tsgc package is demonstrated using data on COVID-19 confirmed cases. |
| Keywords: | Covid-19, Gompertz growth curve, Kalman filter, reproduction number, state space model, stochastic trend, turning points |
| JEL: | C22 I10 C63 |
| Date: | 2024–02–12 |
| URL: | http://d.repec.org/n?u=RePEc:cam:camdae:2407&r=for |
| By: | Roozbeh Qorbanian; Nils L\"ohndorf; David Wozabal |
| Abstract: | Corporate renewable power purchase agreements (PPAs) are long-term contracts that enable companies to source renewable energy without having to develop and operate their own capacities. Typically, producers and consumers agree on a fixed per-unit price at which power is purchased. The value of the PPA to the buyer depends on the so called capture price defined as the difference between this fixed price and the market value of the produced volume during the duration of the contract. To model the capture price, practitioners often use either fundamental or statistical approaches to model future market prices, which both have their inherent limitations. We propose a new approach that blends the logic of fundamental electricity market models with statistical learning techniques. In particular, we use regularized inverse optimization in a quadratic fundamental bottom-up model of the power market to estimate the marginal costs of different technologies as a parametric function of exogenous factors. We compare the out-of-sample performance in forecasting the capture price using market data from three European countries and demonstrate that our approach outperforms established statistical learning benchmarks. We then discuss the case of a photovoltaic plant in Spain to illustrate how to use the model to value a PPA from the buyer's perspective. |
| Date: | 2024–03 |
| URL: | http://d.repec.org/n?u=RePEc:arx:papers:2403.08846&r=for |
| By: | João A. Bastos; Jeanne Paquette |
| Abstract: | Uncertainty quantification associated with real estate appraisal has largely been overlooked in the literature. In this paper, we address this gap by analyzing the uncertainty in automated property valuations using conformal prediction, a distribution-free procedure for constructing prediction intervals with valid coverage in finite samples. Through an empirical study of property prices in the San Francisco Bay Area, we find that prediction intervals obtained using conformal quantile regression have exact coverage. In contrast, prediction intervals obtained from nonconformal quantile regressions severely undercover the data. Furthermore, we show that the intervals adapt to various characteristics of the dwellings, which is crucial given the heterogeneous nature of real estate data. Indeed, we observe that larger and older properties, those in both low and high-income neighborhoods, as well as those on the market for less than one year are more challenging to evaluate. |
| Keywords: | Real estate; Automated valuation model; Conformal prediction; Quantile regression; Machine learning. |
| Date: | 2024–03 |
| URL: | http://d.repec.org/n?u=RePEc:ise:remwps:wp03142024&r=for |