Problem Solving 04

Beka Modebadze 2019 - https://github.com/bexxmodd/econ-papers-reproduction

Replication of a paper published by Kevin L. Kliesen and John A. Tatom - "U.S. Manufacturing and the Importance of International Trade: It’s Not What You Think"

Original paper, published by Federal Reserve Bank of St. Louis Review in 2013, you can download from here and dataset can be downloaded from here

In this session we work on paper that analyzes the factors affecting the growth rate of manufacturing sector in the United

Kliesen, Kevin L., and John A. Tatom. 2013. “U.S. Manufacturing and the Importance of International Trade: It’s Not What You Think” Federal Reserve Bank of St. Louis Review. 95(1), 27-49.

Abstract: The public often gauges the strength of the U.S. economy by the performance of the manufacturing sector, especially by changes in manufacturing employment. When such employment declines, as has been the trend for many years, it is often assumed to be evidence of the slow death of U.S. manufacturing and an associated rise in imports. This article outlines key trends in U.S. manufacturing, especially the strong performance of manufacturing output and productivity, and their connection to both exports and imports. The authors use ordinary regression, causality, and cointegration analyses to provide empirical evidence for the positive role of imports in boosting manufacturing output. Policies to bolster exports at the expense of imports would significantly harm U.S. manufacturing.

Part 1:

The variables and their description are as follows:

Annual Data (Kleiser2013)

Variable Year
ExptoChina Nominal values of U.S. export to China
ExotoWorld Nominal value of total U.S. export
MANValAdded Manufacturing value added. Nominal value.
MANPriceIndex Manufacturing price index (2005=100)
GDPPriceIndex GDP price index (2005=100)
GoodsExp2005 U.S. export of goods, billion dollars, real (2005 chained price index)
GoodsImp2005 U.S. import of goods, billion dollars, real (2005 chained price index)
USGDPReal U.S. real GDP
MANRealInd U.S. Manufacturing production, indexed (2005=100)
DollarNom Nominal value of dollar (trade-weighted)
DollarReal Real value of dollar (trade-weighted)
OilNom Nominal oil price (Refiners’ acquisitions price)
FuelReal Real price of fuel
EquiptSoft Equipment and software fixed private investment
ExpReal Real goods export
ImpReal Real goods import

Our main variable of interest is manufacturing value-added 'MANValueAdded'. It's a time series of values for manufacturing goods produced in the U.S.

To work with time series, we have to set the data as the “time series” and the variable that defines the time series is 'year'

We use manufacturing price index 'MANPriceIndex' to calculate the real value of manufacturing production.

But first, in many cases, the relationship between variables is a logarithmic type of relationship. This means that instead of looking at total values we need to analyze if the percentage change in export/import influenced the percentage change in manufacturing.

To make it clearer, by converting our values into log values we will see if and by how much the change in input (export or import) affects the change of the output (manufacturing)

Now, based on the visual analysis we have to officially confirm the stationarity of the data. We “assume” that the problem is only with stochastic time trends. Regressions on time reveal this type of non-stationarity.

We run regressions of the natural log of real manufacturing on time and check to see if there is a “deterministic trend” by looking at the significance of the coefficient of time. The residual from the below regression is the “detrended data”.

Next to confirm the existance of the time series has a unit root, meaning it is non-stationary we will apply Augmented Dickey-Fuller test. ADF is a one-sided test with left side being the rejection region.

ADF Test Statistics Looks like this:

  • Null Hypothesis {H0}: Data has a unit root (it's non stationary) and variable is time dependent
  • Alternate Hypothesis {H1}: Reject H0; Data has NO unit root; Data is staionary and not time dependant

Now in order to eliminate time's influance on the outcome we need to 'detrend' data to provide the precise and mathematically correct analysis. The residual from the above regression is the “detrended data” we are going to use for further analysis.

Visualization helps again to see what is going on. When we plot residuals, it becomes clear that it has no time trend anymore and the data we have a stochastic process, meaning that plotted residuals follow each other, and we can use. Finally, our data is stationary, and we can start working on it to analyze the effect of exports and imports on manufacturing in the United States.

Part 2:

Assuming there is no other type of non-stationarity, we use the “detrended data” to build a model

First, let’s see if the real manufacturing production follows an AR(1) model

What is Autoregressive Model (AR)?: An AR model predicts future behavior based on past behavior. It’s used for forecasting when there is some correlation between values in a time series and the values that precede and succeed them. You only use past data to model the behavior, hence the name autoregressive (the Greek prefix auto– means “self.” ). The process is basically a linear regression of the data in the current series against one or more past values in the same series. In an AR model, the value of the outcome variable (Y) at some point t in time is — like “regular” linear regression — directly related to the predictor variable (X). Where simple linear regression and AR models differ is that Y is dependent on X and previous values for Y. AR(1) means that we'll use autoregressive model of the first lag.


What is Lag?: A “lag” is a fixed amount of passing time; One set of observations in a time series is plotted (lagged) against a second, later set of data. The kth lag is the time period that happened “k” time points before time i. For example: _Lag1(value in the year 2002) = value in the Year 2001_ and _Lag4(value in the year 2009) = value in the year 2005_. So if the lag is 1, as our model will be, it's called a first-order lag and looks at the value of the previous year.

We see that the past one year transfers its memory by 98.8% to the present. This suggests that we have strong evidence of an AR(1) model being applicable.

We check if higher values of lag can be applicable for our model (if further past then one year can be considered significant enough to add to our model)

# if belove table doesn't works you need to instal this packages !conda install -c conda-forge statsmodels !pip install linearmodels

Part 3:

Next, we would like to take a look at the relationship between manufacturing production and import and export. We work on the “growth rates” of these variables.

First, we find the annual growth rate of manufacturing real value index 'MANRealInd', real import, and real export ('EXPReal' and 'IMPReal'). to calculate the growth rate we will use this formula: alt text Where Yt is the current value and Yt-1 value of the previous year

Now we run a regression of manufacturing growth on export growth

Now we plot imports growth with manufacturing growth

Now we see here a perfect linear trend and the line fits perfectly. This is the evidence that actually with the increase in imports the real manufacturing icnreases

Approx. 1 percent increase in the imports, increases real manufacturing by 0.6%. This finding is similar to what was discovered in the original published paper!

This is an interesting finding. Conventional economics theory suggests that with the ease of the trade wit is expected that the exports (demand for domestic good) increases, thus increases the production (in our case manufacturing). However, we don't see any evidence that an increase in exports has a positive effect on manufacturing. Instead, data shows the evidence that increases in imports increases manufacturing. How so?

Now, this is my take on this case - As the US industry developed over time and became highly skilled and sophisticated. The country started producing more of high skilled artisan goods like jets, military arsenal, robotics, and other technologies and less of the intermediary goods. Here comes the competitive advantage in play. This allowed major US companies and foreign companies to produce intermediate and comparably inferior goods of equal value at a lower price. So other goods which are easy to produce got outsourced, intermediate goods which are used as a part of the production of highly technical and skilled goods got outsourced. So it is hard to tell if we have causation or correlation. Is it that imports stimulated manufacturing growth or manufacturing growth stimulated more imports?