Natural Gas and U.S. Economic Activity.

• Author: Arora, Vipin

1. INTRODUCTION

   Relative to crude oil, comparatively little is known about the impact of the natural gas market on U.S. economic performance. This has become an important issue given recent dynamics in natural gas prices and production. Optimists continually tout the realized and potential economic benefits of the "shale-gas revolution", while many others remain unconvinced about its importance or magnitude. Implicit in either view is an assumption about the past and future macroeconomic impacts of the U.S. natural gas market, but empirical work on which to base this assumption is relatively sparse.

   In this paper we evaluate the importance of the natural gas industry on U.S. macroeconomic performance. Our primary goal is to provide an empirical basis on which to judge the possible impacts of recent developments. We further highlight the channels through which the natural gas market interacts with the U.S. economy, illustrate how this relationship may be changing over time, and consider possible explanations.

   While there are many studies that use macroeconomic models to evaluate the economic importance of natural gas, recent empirical work on the macroeconomic impacts of the U.S. natural gas market is limited. (1) On the empirical side, Kliesen (2006) finds that natural gas prices have historically been unable to predict total U.S. industrial production. This paper uses a single-equation framework and was completed before the large increases in natural gas production due to shale gas. (2) Costello et al. (2006) analyze the interaction between industrial natural gas prices, natural gas consumption, and industrial sector activity and conclude that industrial sector firms respond to relative energy prices, including natural gas prices. However, this paper does not focus on the aggregate impacts of the natural gas market. Similarly, Apergis and Payne (2010) and Sari et al. (2008) find a long-run relationship between natural gas consumption and economic growth in the U.S., but do not consider how the natural gas market can influence economic activity. On the other hand, Weber (2012) considers the economic impacts of the natural gas market after the shale gas boom, but on a regional level. (3)

   The lack of recent empirical work on the relationship between the natural gas market and the U.S. economy is surprising given that natural gas accounts for nearly 25 percent of U.S. energy consumption, making it the second-largest source of energy behind petroleum. (4) Furthermore, consumer expenditures on natural gas have averaged about one percent of personal consumption expenditures (PCE) since 1987, and natural gas is used proportionately between the industrial, commercial and residential, and electric power sectors. (5)

   Given these broad uses for natural gas, there are various ways in which the natural gas market can impact U.S. economic activity. Generally, they can be summarized as working initially through either the supply or demand sides of the economy. The most straightforward supply impact is that changes in the production of natural gas vary output in the oil and gas extraction sector, as well as associated industries. This direct change in production and its ripple through the oil and gas supply chain have been highlighted in many of the recent model-based studies on the economic effects of shale gas (see references in footnote 1 above).

   Natural gas also influences economic activity on the supply-side of the economy through the investment of firms. The application of hydraulic fracturing and horizontal drilling has made a very large resource base available. This potential supply has led to substantial investment in the oil and gas extraction and mining support sectors, as well as other related industries. The expectation that this resource base can support lower natural gas prices for a sustained period is also leading to investment by firms outside of the oil and gas industry which rely on natural gas as an input.

   Lower natural gas prices, irrespective of their cause, lower input costs for firms. These can be passed on to consumers by allowing firms to supply the same amount of goods and services at lower prices. Firms may also realize higher profits, which can lead to additional hiring, capital investment, higher dividends, or saving. Each has a follow-on impact on the economy-wide demand for goods and services. Lower prices also directly influence demand through consumers. They can raise disposable income, lower precautionary savings (or raise it in the case of a price rise), or cause consumers to change their plans for the purchases of durable goods (Kilian, 2008). Each of these results in alterations to the economy-wide demand for other goods and services stemming from the initial variation in the price of natural gas.

   With these various channels in mind, we estimate a structural vector autoregression (VAR) to assess the impact of the natural gas market on U.S. economic activity. Our monthly four-variable model characterizes the supply, demand, and price of U.S. natural gas. The results are presented through impulse response analysis and variance decompositions of the model's forecast error. We also conduct sensitivity analysis on our results, including consideration of how recent developments in the U.S. natural gas market may be affecting the VAR.

   The VAR model and related sensitivity analysis lead to two primary conclusions. The first is that natural gas supply changes are the primary means through which the U.S. natural gas market impacts domestic economic activity. Variations in natural gas demand for heating and power or other factors, while important for the natural gas price, do not impact economic activity in a substantial way. Our second conclusion is that the shale gas revolution has in fact changed the relationship between natural gas supply and U.S. economic activity. The responses of industrial production to the same increase in natural gas supply are larger after 2008 than before, although the size of this change remains unclear at this point.

2. VAR MODEL

   A general VAR process can be encapsulated by a mean-zero moving average representation, without any deterministic terms (Lutkepohl, 2007):

   [mathematical expression not reproducible] (1)

   where y is an N x 1 vector of variables, the [B.sub.j] are N x N matrices of coefficients, and the reduced form errors ([u.sub.t]) are N x 1 white noise processes with E([u.sub.t][u.sub.t.sup']) = [S.sub.u]. The coefficient matrices ([B.sub.j]) summarize the responses of the variables to the respective errors. Because [S.sub.u] is not necessarily diagonal, the errors may be correlated across equations in the same time period. As is well-known, this can make interpretation of any responses misleading, because co-movement with other variables is not taken into account.

   An equivalent representation of the moving average process with orthogonal innovations can circumvent this issue. In this case the transformed innovations will be uncorrelated by construction, so that the variance-covariance matrix of the shocks is diagonal. The identity matrix is often chosen in this case, which amounts to finding an N x N matrix G such that:

   G[S.sub.u]G' = I (2)

   where I is the N x N identity matrix. The orthogonal innovations are [[epsilon].sub.t] = G[u.sub.t], so that E([[sigma].sub.t][[sigma].sub.t.sup']) = GE([u.sub.t][[epsilon].sub.t.sup'])G = I. These innovations are uncorrelated across both time and equations. In this case equation (1) can be rewritten:

   [mathematical expression not reproducible] (3)

   In this equation the [B.sub.j]G summarize the impulse responses which are plotted below. Also j used in subsequent analysis to summarize model results are forecast error variance decompositions.

   The forecast error variance decomposition can be reconstructed by recognizing that [mathematical expression not reproducible] so that the error of the i-step ahead forecast is (Enders, 2010):

   [mathematical expression not reproducible] (4)

   From this equation we can extract the total variance in the i-step ahead forecast error of variable j, as well as the variance in the error of variable j due to variable k. The variance...