4. Empirical Results
4.1. Ordinary Least Square Regressions
4.1.2. CEPs Model 2— Diesel Prices, DXY, and Inventory Over Sales
國
立 政 治 大 學
‧
N a
tio na
l C h engchi U ni ve rs it y
4.1.2. CEPs Model 2— Diesel Prices, DXY, and Inventory Over Sales
Next, a VAR using macroeconomic data on commodity prices was conducted. The key variables to be used as dependent variables included CEPs, diesel prices, and the DXY. All variables were one-year differenced. These particular variables were chosen for the model for several reasons, specifically from what the literature had suggested is correlative with capital prices. Theoretical foundations were also taken into consideration. Moreover, one-month differenced iron and steel prices were used since one-year differenced variables failed to pass the Dickey Fuller unit root test hurdle.
After these tests were completed, the selection order criteria tests were conducted. These tests found that the statistically significant lag order was 10 for the LR, FPE, and AIC selection order criteria statistics. Importantly and as mentioned within the methodology section, the AIC statistic is one of the m ore important selection order criteria to be used for its reliability in not over-estimating the VAR model and not underestimating it as well. Because of this, a model with 10 lags was used for the VAR model.
‧
國立 政 治 大 學
‧
N a
tio na
l C h engchi U ni ve rs it y
Table 4-8 Lag Order Selection Criteria Tests for the DXY, Diesel Prices, and the Ratio of Inventory Over Sales for Manufacturers
The VAR was then conducted. These statistics show that all variables according to a chi-squared distribution are statistically significant (see Table 4-9). Moreover, the R-chi-squared statistic for all the variables was substantively high. This is true for all variables with the exception of iron and steel prices differenced by one month.
Table 4-9 VAR Equation Results for the DXY, Diesel Prices, and the Ratio of Inventory Over Sales for Manufacturers
After the VAR equations were conducted, the OIRFs were generated. Again, we focus only on the relationship between these variables as the independent variables and CEPs as the dependent
‧
國立 政 治 大 學
‧
N a
tio na
l C h engchi U ni ve rs it y
variable. As can be seen, a shock of diesel prices increases the moving-average of CEPs in the first seven steps after the shock. After this, there is a 90 percent chance that CEPs are positive for around another seven steps (see Figure 4-7). Throughout the entire duration except for the first few steps, the moving average response of CEPs was positive with the exception of slightly below in the first step and slightly below in the 20th step.
Figure 4-7 OIRF of Diesel Pries Impulse and CEPs Response
The next OIRF to be generated was that of one-month iron and steel prices on one-year CEPs. The OIRF results assert that there was a 90 percent chance CEPs were negative in the first period after a shock of iron and steel prices. This was likely due to the speculative nature of commodity markets and a reminder that correlation does not equal causation. This should be kept
‧
國立 政 治 大 學
‧
N a
tio na
l C h engchi U ni ve rs it y
in mind primarily due to the results after the first step in which three steps and beyond the moving average is entirely positive. Moreover, there was a 90 percent chance that this impact was entirely positive for a number of steps from around the 8th until the 15th step. Because of this, it can be asserted that there is a generally a positive impact from short-term iron and steel prices on long-term CEPs.
Figure 4-8 OIRF of Iron and Steel Pries One-Month Differenced Impulse and CEPs One-Year Differenced Response
The next OIRF used the one-year differenced DXY against the one-year differenced CEPs.
As can be seen, the DXY during the first periods is positive with a 90 percent certainty of being positive from around the third to fifth period while there is then a 90 percent certainty that it will
‧
國立 政 治 大 學
‧
N a
tio na
l C h engchi U ni ve rs it y
become negative starting at around the tenth period until the 20th period. This is likely due to the notion that in the short-run, prices increase on equipment since it is an inelastic marketplace. This also means that equipment orders still go through even though the dollar may be moving higher against other currencies. It therefore takes a significant amount of time (around a year according to this data) for the stronger dollar to translate into more purchases of foreign machinery. After time has passed, purchasers of machinery are more adept at purchasing machinery from abroad with their strong US dollar and domestic competitors are forced to lower their prices.
‧
國立 政 治 大 學
‧
N a
tio na
l C h engchi U ni ve rs it y
Figure 4-9 OIRF of the DXY Impulse and CEPs Response
The next test used the Granger causality equation to determine the directionality of CEPs and commodity variables against one another. The results show that CEPs do not Granger cause any commodity variables. However, the results also show that the DXY and one-month differenced iron and steel prices Granger caused CEPs. Moreover, there were other relationships of Granger causation between the commodities. This included the Granger causation of diesel prices on the DXY, one-month differenced iron and steel prices on diesel prices, and diesel prices on one-month differenced iron and steel prices.
‧
國立 政 治 大 學
‧
N a
tio na
l C h engchi U ni ve rs it y
Table 4-10 Lagrange-Multiplier Test for CEPs Model 2
The Lagrange-multiplier test was then conducted after the VAR. As can be seen, the first, second, eight, and tenth lag were found to not be statistically significant. Therefore, these lags lacked autocorrelation. Since the lags which did reject the null hypothesis were all of a lower order than the largest lag, this parameterization was accepted again to avoid overparameterization given the other parameters which were placed upon the model such as lag order selection criteria.
‧
國立 政 治 大 學
‧
N a
tio na
l C h engchi U ni ve rs it y
Table 4-11 VAR Granger Causality Results for the DXY, Diesel Prices, and the Ratio of Inventory Over Sales for Manufacturers
In this Granger causality matrix, it appears to be the case that iron and steel prices differenced by one month are much more impacted by the range of variables as opposed to the other way around. This is also the case for CEPs. In this model, on the DXY Granger causes CEPs while there are a number of other interactions in the model which account for price action.
‧
國立 政 治 大 學
‧
N a
tio na
l C h engchi U ni ve rs it y
Figure 4-10 Granger Causality Map for the DXY, Diesel Prices, and the Ratio of Inventory Over Sales for Manufacturers
The final test used with this particular model were the dynamic forecasts of the variables into the year 2021. Similarly to the previous dynamic forecast which looked at technology variables, CEPs enjoyed a 90 percent chance of decreasing into 2020 and nearly enjoy a 90 percent chance of increasing throughout 2020 into 2021. However, the 90 percent confidence interval is not entirely above the 0 line so therefore this determination is not statistically sound. Moreover, the impact of the DXY is not certain either as the 90 percent confidence band hugs the 0 line throughout 2020 and into 2021. One-month differenced iron and steel prices also suffered a similar fate. As for diesel prices, there was a 90 percent chance that there was a negative drop followed by a positive increase at the end of 2020, but these confidence bands also suffered a similar fate of the DXY and iron and steel prices. Because of this, further specification would likely be required
‧
國立 政 治 大 學
‧
N a
tio na
l C h engchi U ni ve rs it y
in order to create more accurate dynamic forecasts. This likely includes increase the lag number in the VAR model.