Sampling the Future: A Bayesian Approach to Forecasting From Univariate Time Series Models

Abstract

The Box-Jenkins methodology for modeling and forecasting from univariate time series models has long been considered a standard to which other forecasting techniques have been compared. To a Bayesian statistician, however, the method lacks an important facet-a provision for modeling uncertainty about parameter estimates. We present a technique called sampling the future for including this feature in both the estimation and forecasting stages. Although it is relatively easy to use Bayesian methods to estimate the parameters in an autoregressive integrated moving average (ARIMA) model, there are severe difficulties in producing forecasts from such a model. The multiperiod predictive density does not have a convenient closed form, so approximations are needed. In this article, exact Bayesian forecasting is approximated by simulating the joint predictive distribution. First, parameter sets are randomly generated from the joint posterior distribution. These are then used to simulate future paths of the time series. This bundle of many possible realizations is used to project the future in several ways. Highest probability forecast regions are formed and portrayed with computer graphics. The predictive density's shape is explored. Finaliy, we discuss a method that allows the analyst to subjectively modify the posterior distribution on the parameters and produce alternate forecasts.

A Hybrid Least Squares Support Vector Machine With Bat and Cuckoo Search Algorithms for Time Series Forecasting

“…Univariate statistical approach predicts y from trends alone, while multivariate statistical approach predicts y from trends and other variables. The univariate approach includes the use of filtering techniques such as Kalman filtering (Louka et al, 2008) and exponential filtering (Chan, Dillon, Singh, & Chang, 2012) whereas Autoregressive Integrated Moving Average (Lee & Tong, 2011), K-nearest neighbor (Fan, Guo, Zheng, & Hong, 2019) and Bayesian (Thompson & Miller, 1986) are examples of the multivariate approach. Despite the statistical approach yielding acceptable estimates, they do not address the nonlinear characteristics of forecasting (Chan et al, 2013).…”

Uncertainty and Density Forecasts of Arma Models: Comparison of Asymptotic, Bayesian, and Bootstrap Procedures

“…(), and Ehlers and Brooks (). Thompson and Miller () overcome some of the computational difficulties allowing for other error distributions.…”

“…The Bayesian procedure of Thompson and Miller () is illustrated by implementing it to construct forecast intervals for the AR(2) and ARMA(1,1) models considered previously. When the Bayesian procedure is implemented assuming Gaussian (Student‐ν) forecast errors, it is denoted as BAYESN (BAYEST) .…”

Forecast Uncertainty, Its Representation and Evaluation

“…These can be reported in a table, or as yet further lines on a time series plot of forecasts for several periods into the future, generalising the plot of an interval forecast described above. With selective shading of quantiles "to draw attention away from point forecasts and toward the uncertainty in forecasts" this was first proposed by Thompson and Miller (1986). Since the dispersion of the distribution increases and the intervals "fan out" as the forecast horizon increases, such plots have subsequently become known as "fan charts".…”