Forecasting Natural Gas Prices Using Time Series Models

Abstract

The objective of this thesis is to estimate the natural gas component of the All Urban Consumer Price Index (CP-U) using time series forecasting models. Being able to accurately predict future CPI-U values is important because it allows portfolio managers and financial institutions to properly adjust their market positions for inflation prior to the CPIs print. Certain bonds, such as Treasury Inflation Protected Securities (TIPS), are indexed to CPI and accrete in value as the CPI increases. Accurately knowing the print of the CPI prior to its publication allows investors to properly anticipate movements in TIPS. Data for this thesis were gathered from two sources: daily Henry Hub natural gas spot prices retrieved from Bloomberg and monthly Bureau of Labor Statistic (BLS) data on CPI-U values for natural gas. Each of these data series was tested for unit root processes, seasonality, and cointegration. Once these two data sets were shown to be non season, cointegrated series, they were run through three models, which were evaluated in terms of fit and performance in and out of sample against one another. The model which proved to have the greatest predictive power was Model 3, in which a distributed lag polynomial was fit to the data and used to create a rolling out of sample regression.