3. Methodology
3.1. Descriptive Statistics and Properties of Variables
excluding the time sensitive variables after which Model 9 includes the time sensitive variables.
Model 10 technological variables and time variables and Model 11 excludes technological variables and commodity variables while including both time variables. The reason why Model 10 and Model 11 do not follow the same pattern as Model 1 through Model 9 is due to the fact that results did not change between Model 10 and a model which included time variables. This model was excluded in the results because it did not add any value to the analytical process since the results were the same as Model 10. The same was the case for a model similar to Model 11 which did not include time sensitive variables and so it too was excluded while Model 11 with time sensitive variables was included. The same exact pattern for Regression Table 1 is used for Regression Table 2. The reason the same model is not used for Regression Table 3 is because the dependent variable in the third series of regressions is already time sensitive in that the inclusion of variables within different models does not change the number of observations in the regressions.
3.1. Descriptive Statistics and Properties of Variables
The different models using the ordinary least square method is constant in that all independent variables are the same, again though with the last model using one-month differenced data. The rationalization of the framework is deeply embedded within the previous work conducted in the academic literature. Since most of the literature does not directly address correlates of CEPs, the model used to determine the correlates of CEPs is prefaced on the literature surrounding the correlates of equipment prices in general. Primarily, this based upon the propensity for the literature to focus on technology, macroeconomic indicators, and the inputs associated with the cost of production.
‧
Table 3-1 One-year Differenced Variable Codes and Descriptions
Variable Code Variable Description
equipppi Construction equipment prices (CEPs)
excadargcr~e Excavator, crane, and dragline prices (ECDPs)
generalmac~y General machinery prices (GMPs)
nasdaq The NASDAQ Composite
dataproces~g Data processing CPI
rdemplyscali R&D Employees in California
effectivf~s Effective federal funds rate
recesdummy Recession dummy
techcpi Technological hardware CPI
ironsteel Iron and steel CPI
ratioinven~s Ratio of inventory over sales for manufacturers
ipbusequip Industrial production of business equipment
gdp Gross domestic product (GDP)
diesel Diesel CPI
dxy Trade weighted dollar index
nonrescons~g Non-residential construction spending
There are three primary groups of independent variables. The first group of variables that was regressed included technological variables. These included the NASDAQ Composite, R&D employment in California, technological hardware CPI, and data processing CPI. The NASDAQ Composite acts as a proxy for technological investment. R&D employment acts as an important
‧
國立 政 治 大 學
‧
N a
tio na
l C h engchi U ni ve rs it y
variable in quantifying the degree to which technological advancement is being pursued beyond monetary expenditures. In other words, it can be used to quantify or as a proxy of the abstract concept of human capital or investment in human capital, both of which plays key roles in technological advancement while also suffering from the difficulty of quantification.
Technological hardware CPI simply quantifies the cost of the materials used to produce technological equipment. Finally, data processing CPI is also another variable which attempts to quantify the impact of human capital on equipment prices through human capital and technological advancement.
Table 3-2 Descriptive Statistics of Ordinary Least Square Regression Variables
The second set of variables in the model are all associated in some form or another with in general macroeconomic indicators, supply and demand, monetary policy, and overall economic conditions. These include the effective federal funds rate, the ratio of equipment inventory over equipment sales by private companies, the industrial production of business equipment, the OECD’s monthly measurement of US GDP, nonresidential construction spending, and a US recession dummy variable.
‧
The effective federal funds rate is not only a fundamental component to monetary policy and can proxy as the government’s impact on market forces, but interest rates are also crucial in borrowing costs, investment expenditures, and acts as the cost of capital (equipment prices). The ratio of equipment inventory over equipment sales, while not a traditional measurement, provides information on the demand of construction equipment and the speculative appetite of commercial actors who purchase goods from producers. Since companies that purchase equipment directly from producers tend to sell this equipment to contractors or other companies which intend to use equipment, the ratio of inventory to sales of provides insights on the relationship between speculative demand and market demand in reference to the price in which the equipment is being quoted. The industrial production of business equipment quite simply measures the supply of construction equipment. GDP, while having been quoted as being correlated with CEPs in the academic literature, also likely impacts the price of construction equipment, but in a counterintuitive way given the lagging nature of equipment prices and the state of the business cycle. Nonresidential construction spending is chosen over all construction spending since residential construction is much more speculative in nature over the past few decades in the US, does not reflect the demand of the construction market, and suffered a dramatic bubble in 2008 which was the result of dramatic market distortions. Finally, a recession dummy is used to test if any variables were spurious and also to determine what the direct relationship between equipment prices and the business cycle is.
‧
國立 政 治 大 學
‧
N a
tio na
l C h engchi U ni ve rs it y
Table 3-3 Correlation Coefficient Matrix of One-Year Differenced OLS and VAR Regression Variables
‧
Manufacturing unemployment is held constant primarily due to the fact that machines are primarily not produced by individuals, but are rather produced by other machines. Because of this, employment in the manufacturing industry likely plays a very small role in the price of the production of construction equipment in the United States. Taking this factor into consideration, the model is based upon the more recent impact of technological progress which has over the past few decades increased at greater speeds than prior years past. Moreover, the regressions use a form of employment, R&D employment, which is strongly and positively correlated with other forms of employment. Therefore, another employment figure or proxy variable more directly related to the production of construction equipment would have been redundant and unnecessary.
One question that arises within the variance of the variables which will be seen in the results section is the issue of why do financial markets impact equipment prices so consistently and so strongly compared to other variables. This is especially the case for the NASDAQ Composite which enjoyed a strong negative relationship with equipment prices of all kinds in the OLS models, but also enjoyed strong relationships with all types of equipment prices in in the VAR models. On the other hand, commodity prices such as the DXY, diesel prices, and iron and steel prices also were shown to have very strong and consistent impacts on equipment prices (mostly positive except for the DXY which was positive in the short-term and negative in the long-term).
Why is this the case? First of all, it may be due to the nature of the data and the impact of transformation. Financial markets tend to be much more volatile than macroeconomic data points.
Since financial markets are constantly spitting out new data points every second with data points sometimes measured in less than one second increments, these markets can suffer from dramatic outliers. Oftentimes, these outliers correct themselves and are retraced rather quickly. A good example of the rapidity of financial markets is what is known as flash crashes where a financial
‧
國立 政 治 大 學
‧
N a
tio na
l C h engchi U ni ve rs it y
instrument can drop by an extremely large amount, but recover in a matter of minutes or seconds occurring in a ‘flash’. Even in the event that flash crashes do not occur, financial markets at a daily or weekly level can change dramatically. This inevitably means there may be more extreme examples of outliers.
At any rate, the leptokurtotic nature of differenced variables is more pronounced in financial markets as can be seen in Figure 3-1. The first histogram of CEPs clearly shows a more plateaued distribution. While the two variables have a similar mean (3.8 for CEPs and 4.5 for iron and steel prices), but the two variables have incredibly different standard deviations with equipment prices being 2.99 and iron and steel prices as 26. The highly leptokurtotic variables are much more normally distributed which usually translates into the residuals of OLS and VAR models are much more normally distributed. Finally, the histograms show us just how leptokurtotic the financial market variables tend to be with iron and steel prices enjoying main bin consisting of more than 100 observations while he main bin of equipment prices is less than close to 70.
Figure 3-1 Histogram for Mesokurtic Construction Equipment Prices and
Leptokurtic Iron and Steel Prices
‧
means that it suffers from a statistical property which contradictorily makes it a better predictor variable. Since all random walk processes enjoy a normal distribution, we can measure the degree to which variables are normally distributed and therefore the degree to which variables are stochastic. Normally distributed variables do not suffer from high kurtosis which is generally considered to be around 5, but anything above 3 is larger than a perfectly normally distributed variable. Any variable with a kurtosis of larger than 5 is considered to suffer from leptokurtosis and is leptokurtic. Moreover, a large kurtosis indicates the data suffers from large outliers.However, it is also important to remember the distinction between statistical outliers and numerical outliers. Leptokurtosis translates into higher volatility because there are so many values which are closer to the mean. Therefore, the standard deviation is much smaller in leptokurtotic variables.
As far as financial data goes, the characteristic of leptokurtosis is a “stylized fact” of financial markets and is therefore unavoidable when modeling data from stocks or commodities (Alexander, 2009, pg. 106). It may also be the case that leptokurtotic variables increase the likelihood of Type I errors (Ibid.). However, there is also evidence to suggest that leptokurtotic datasets do not increase Type I errors (Ocampo and Rodriguez, 2012) suggesting that there is no impact on null hypotheses testing at all.