4. Empirical Results
4.2. ECDP VAR Models
4.2.1. ECDP Model 1—Data Processing, the NASDAQ, and Technology Hardware CPI
國
立 政 治 大 學
‧
N a
tio na
l C h engchi U ni ve rs it y
4.2.1. ECDP Model 1—Data Processing, the NASDAQ, and Technology Hardware CPI
The first VAR for ECDPs and include data processing CPI, the NASDAQ Composite, technology hardware CPI, and ECDPs. First, the results of the Dickey Fuller unit root test are required for the ECDPs followed by the output statistics of the VAR and the orthogonalized version of IRFs of each variable against the equipment prices. A total of 12 lags were used for the whole model. Such was the case for the first and second VAR that included the ECDPs.
Table 4-20 Lag Order Selection Criteria Tests for ECDPs, the NASDAQ, Data Processing CPI, and Technology Hardware CPI
The above figure shows that the most consistently statistically significant lag order is 6 lags. Again, the AIC suggests this is the case which is the most heavily weighted statistic in determining lag order throughout this thesis. Moreover, the FPE statistic also suggests 6 lags is the best fit for the model. Because of this, 6 lags were used for this current model.
‧
國立 政 治 大 學
‧
N a
tio na
l C h engchi U ni ve rs it y
Table 4-21 VAR Equation Results for ECDPs, the NASDAQ Composite, Data Processing CPI, and Technology Hardware CPI
The results of the VAR equations suggest that all variables are statistically significant (see Table 4-21). Moreover, all variables enjoyed a relatively high R-squared statistic suggesting that the variance in the sample size strongly reflects the variance in the population. The number of observations from these VARs, it should be remembered, is much smaller than the number of observations with CEPs. However, for the current VAR, this is a non-issue since the inclusion of data processing would reduce the sample size of a VAR including CEPs anyway. At any rate, the other VARs with ECDPsis about 33 percent smaller than the other VARs.
‧
國立 政 治 大 學
‧
N a
tio na
l C h engchi U ni ve rs it y
Figure 4-21 OIRF of One-Year Differenced Technology Hardware CPI Impulse and ECDPs Response
The first OIRF (above) uses technology CPI as the impulse variable and equipment prices as the response variable (see Figure 4-21). As can be seen, there is a 90 chance that the impact of an increase in technology CPI negatively impacts equipment prices. Moreover, this is the case for nearly all of the first ten steps after the initial ‘shock’. With a strong certainty, it can be asserted that technology CPI enjoys a negative relationship with ECDPs.
‧
國立 政 治 大 學
‧
N a
tio na
l C h engchi U ni ve rs it y
Figure 4-22 OIRF of One-Year Differenced NASDAQ Composite and ECDPs Response
The relationship between the NASDAQ Composite and these particular equipment prices though is much less certain (see Figure 4-22). While the relationship appears to be negative, there is not a 90 percent confidence that this relationship is certainly negative after the initial ‘shock’
since the 90 percent confidence intervals hug the zero-reference line throughout the duration of the steps. While this is somewhat disconcerting, the other methodologies used throughout the thesis reinforce the notion that this may be an over or under parameterized VAR which could be the reason behind the weak results. If one were forced to give an opinion though on the relationship between these two variables though, it may be more accurate to take into consideration the OLS
‧
國立 政 治 大 學
‧
N a
tio na
l C h engchi U ni ve rs it y
linear regression results which asserts a more confident negative relationship between these two variables when time is held constant. Overall, the relationship appears to be negative between the NASDAQ and ECDPs.
Figure 4-23 OIRF of One-Year Differenced Data Processing CPI Impulse and ECDPs Response
The next OIRF uses data processing CPI as the impulse variable with equipment prices as the response variable (see Figure 4-23). While the relationship appears to be negative, the 90 percent confidence intervals are only both negative for a handful of steps or around five.
Nonetheless, moving average of the OIRF is entirely negative for all 20 steps. Because of this, as with previous uncertain results, it may be more conservative to refer to the OLS linear regression
‧
國立 政 治 大 學
‧
N a
tio na
l C h engchi U ni ve rs it y
results of the relationship between these two variables. When holding time constant as does the OLS linear regression, then it appears to be the case that there is a negative relationship which is consistent with the moving average results of the OIRF generated from the VAR.
Figure 4-24 OIRF of One-Year Differenced ECDPs Impulse and ECDPs Response
The last OIRF simply examines the relationship between equipment prices against equipment prices. The impulse and response variables are both ECDPs (see Figure 4-24). The results suggest that it takes at least 7 steps from a shock of prices on themselves to return to equilibrium level. This is a noticeably shorter time than that of CEPs which is likely due to the fact that these prices are much less volatile or perhaps speculative than CEPs.
‧
國立 政 治 大 學
‧
N a
tio na
l C h engchi U ni ve rs it y
Table 4-22 Granger Causality Test Results for ECDPs, the NASDAQ Composite, Data Processing CPI, and Technology Hardware CPI
A Lagrange multiplier test for model validity was then conducted. As can be seen in Table 4-23, only the third and fourth lag rejected the null hypothesis that no autocorrelation was present at the lag order. While not all lags lack autocorrelation, it is important to not overparameterize the model so as to eliminate the issue of preventing models from being created that could provide relatively valid results. Because of this consideration and the statistical fact that four of the six lags lack autocorrelation, the VAR is considered relatively valid.
‧
國立 政 治 大 學
‧
N a
tio na
l C h engchi U ni ve rs it y
Table 4-23 Lagrange-Multiplier Test for ECDPs VAR Model 1
After the Lagrange multiplier tests were conducted, dynamic forecasts were created based on the results of the above VAR. The results are quite promising compared to the dynamic forecasts where CEPs were used; all of the dynamic forecasts are much more precise in their ability to predict whether or not the figures are positive or negative over a long-period of time. This suggests that the variable of excavators, cranes, and dragline prices are a better fitted variable to use in modeling for equipment prices than what is the in the general category of CEPs. However, this is not the case for the price of equipment itself which suffers from more uncertainty in that it is 90 percent likely to be positive throughout the end of 2019. Throughout 2020 until the last few months of the forecast though, the 90 percent confidence band is above and below the zero-point line.
In spite of this, there is 90 percent certainty that the NASDAQ Composite will be positive for the duration of 2020 while technology CPI will continue to decrease. Moreover, it is forecasted
‧
國立 政 治 大 學
‧
N a
tio na
l C h engchi U ni ve rs it y
that the price of data processing will increase throughout the duration of 20202 into 2021. These conflicting trends are likely the culprit of the uncertainty surrounding equipment prices. This is the case since higher NASDAQ prices means lower equipment prices and so too does lower technology CPI prices. However, higher data processing prices means higher equipment prices.
Therefore, these conflicting trends create more uncertainty in the medium-term if one were entirely to rely upon only this forecast and only this model, the adoption of which would be ill-advised.