**1. Introduction**

Researchers have studied crude oil for many years because it is of major interest as a significant but limited resource. In addition, the crude oil index plays an important role among economists, policymakers, and investors, because fluctuations in either its price or its production have a remarkable impact on the world economy and stock markets. On the contrary, as a relatively environmentally friendly resource and a key fuel for the electrical power and industry sectors, natural gas has attracted increasing attention. Evidence for this increasing attraction comes from the fact that the consumption of natural gas has risen to a share of 23% of total energy resources, and is the fastest-growing fossil fuel among energy resources [1]. Another fact is that the demand for natural gas increased by 4.6% in 2018, accounting for almost half of all energy resource demand growth [2]. Thus, it is not difficult to conclude that fluctuations in the prices of crude oil and natural gas greatly influence economic activity and daily life.

Researches had shown that there is a correlation between US economic growth and exogenous oil shocks. For instance, Hamilton [3,4] and Hooker [5] proposed that after oil price increases, there were recessions in the US economy. However, as proposed by Kilian [6], depending on the underlying causes, such as supply shocks, aggregate demand, or precautionary demand, price fluctuations have varying dynamic effects on real prices, and hence on the economy. Other studies have agreed; for instance, Kling [7] reported that the association is obvious between crude oil price increases and declines in the stock market. However, Kilian and Park [8] reported that US real stock returns reacted differently to oil price shocks, depending on the origins driving the oil price shocks. Other research has also

analyzed the macroeconomic e ffects of oil market fluctuations. Oladosu et al. [9] adopted a quantitative meta-regression model to simulate the oil price elasticity of GDP in US and their results revealed that the estimated US GDP elasticity was negative, and particularly smaller than about 10 years ago. Jan van de Ven and Fouquet [10] identified the impact of energy shocks on economic activity using data from the United Kingdom for 310 years and their study indicated that the influences of supply shocks declined due to the partial shift from coal to crude oil, and that it is possible to reduce vulnerability and increase resilience by substituting renewable energy sources for traditional energy sources. Ju et al. [11] used data from 1980 to 2014 to investigate the macroeconomic performance of oil price shocks by adopting three methods, namely, empirical covariance, robust covariance, and support vector machine methods, and the results implied that the outlier performances of GDP, CPI, and the unemployment rate are consistent with the oil price shock process. Correspondingly, some literature investigated the relationship between natural gas price shocks and macroeconomic aggregates. Nick and Thoenes [12] extended the structural vector autoregression (SVAR) model and analyzed the e ffects of di fferent fundamental influences on the price of natural gas in the German market. Their results revealed that in the short-term, the price of natural gas tends to be a ffected by some factors, such as temperature, inventory, and supply insu fficiencies, and, by contrast, in the long-term the price tends to be influenced by the overall economic climate and the substitutional relationship between crude oil and coal. Zhang et al. [13] used a computable general equilibrium model to investigate the macroeconomic e ffects of natural gas prices in China and their results showed that when natural gas prices increase, the CPI would increase and GDP would decrease. In addition, research has also emphasized the transmission effect of oil demand and supply shocks on the natural gas market. For example, Jadidzadeh and Serletis [14] analyzed the e ffects of supply-demand shocks stemming from the global oil market on the real price of natural gas and found that oil supply-demand shocks accounted for approximately 45% of the fluctuation in the price of natural gas.

This study's first objective is to decompose shocks to the real price of crude oil and natural gas into three components: (1) oil/gas supply shocks, (2) global aggregate demand shocks, and (3) specific or precautionary demand shocks. Its second objective is to evaluate important di fferences in how US macroeconomic aggregates, such as CPI and real GDP growth, react to various oil shocks underlying the real price of crude oil and natural gas. The third objective of this study is to compare its results with Kilian's results regarding crude oil in 2009, in order to determine whether conditions have changed since 2007. Our data cover the period from 1973:1 to 2019:6, whereas Kilian [6] only had data up to 2007:12. Its fourth objective is to compare the impacts on crude oil and natural gas of di fferent types of shocks to US macroeconomic aggregates. We expect this study to have implications for market operators and investors.

Because one of the objectives of our study is to compare our results with Kilian's work, and because we used the same approach as Kilian [6], there is a necessity to demonstrate the similarities and discrepancies between the two works. We used the same methodology and variables as in Kilian's work. However, because the importance and the potentiality of natural gas has gradually increased, we added natural gas to our study to check the e ffects of each structural shock on the real price of oil and gas and compare their e ffects on the US macroeconomic aggregates; in contrast, Kilian's work focused on the crude oil market. Another di fference is that our sample period of crude oil is longer than Kilian's to verify if there are any changes after 2007.

Our findings in this study are threefold. First, by constructing an SVAR model to quantify the responses to one-standard-deviation structural shocks, we found that there were di fferences between the e ffects on crude oil and natural gas in the response patterns of real economic activity due to precautionary demand shocks. Second, by decomposing historic real prices to check the cumulative contribution from each demand and supply shock to the real price, we found that the cumulative e ffect of precautionary demand shocks was varying in degree between crude oil and natural gas; specifically, the cumulative contribution of an oil-specific demand shock to the real oil prices is larger than the cumulative contribution of a gas-specific demand shock to the real gas prices. Third, by utilizing the

regression model to investigate how oil and gas demand and supply shocks that underlie the real price of oil and gas influence US macroeconomic aggregates, we found that precautionary demand shocks on crude oil and natural gas have di fferent e ffects on CPI inflation and had similar e ffects on real GDP growth. Both oil- and gas-specific demand shocks led to a small, but statistically insignificant, reduction of the US GDP level. Oil-specific demand shocks tended to cause a large and statistically significant increase in CPI inflation; by contrast, gas market-specific demand shocks led to a small and statistically insignificant increase in the US CPI level. This result implied policymakers should react di fferently to gas-specific demand shocks and oil-specific demand shocks.

The remainder of this paper is organized as follows. Section 2 provides a detailed description of the data and introduces the econometric model, a two-stage method based on the SVAR model proposed by Kilian [6]. Section 3 identifies the structural shocks that drive the real price of crude oil and natural gas. We quantify the historical evolution of these shocks and the response to these shocks from production, real activity, and the real price of crude oil and natural gas. We decompose the real price of crude oil and natural gas over time to assess the respective cumulative contribution of each shock to real prices. Finally, we analyze the e ffects of those shocks on US macroeconomic aggregates. Section 4 concludes the study.

#### **2. Materials and Methods**

## *2.1. Data*

Table 1 describes the datasets we utilized and their sources. These datasets included crude oil production, gas production, a real economic activity index (the Kilian index available on his website), crude oil prices, natural gas prices, and US real GDP, deflated using US CPI (Kilian [6]).


**Table 1.** Data descriptions and sources.

1 Real price means price which is deflated by US CPI. 2 Data from the International Energy Agency.

We collected data from Bloomberg, Thomson Reuters, and OECD Stat. These datasets include the crude oil spot price (West Texas Intermediate, WTI) and the natural gas spot price (Henry Hub). *l*\_*rpot* defers to the real price of crude oil expressed in log terms. *l*\_*rpgt* defers to the real price of natural gas expressed in log terms. Regarding the dataset of the Kilian index, we collected the data from the Lutz Kilian website. From Kilian (2009), the Kilian index is a detrended real freight rate index constructed as a measure of monthly global real economic activity, which could imply the variation of worldwide real economic activity and capture the shifts in the demand for industrial commodities throughout international markets [6,15] (see Appendix A for details). In this study, *l*\_*reat* defers to the Kilian index; because the minimum of the Kilian index is −161.643, we added 196 to every data point and expressed the series in log terms. We collected production data for oil and gas from the International Energy Agency (IEA), and calculated the change rate of oil/gas production in percentage terms to reflect the percentage change of worldwide oil/gas production. <sup>Δ</sup>*oprodt* refers to the percentage change in global crude oil production, and <sup>Δ</sup>*gprodt* refers to the percentage change in global natural gas production. US

real GDP growth and US CPI data was collected from the Organization for Economic Co-operation and Development (OECD); because data for the US real GDP growth rate is unavailable at a monthly frequency, we adopted the quarterly frequency when we downloaded the US real GDP growth rate and US CPI datasets. The sample period for crude oil data was from January 1973 to June 2019, but the natural gas sample was limited by the available data period of January 1994 to June 2019.

The descriptive statistics and stationarity tests for the variables are shown in Table 2. This table indicates the mean, maximum, minimum, and standard deviation of each variable. (Although the log of the real price of natural gas may have a unit root, we checked the stability condition of the SVAR system and found that the system is stationary; more details are provided in Appendix B.)


**Table 2.** The results of unit root tests for the variables.
