Chapter 14 Replicating ADH

14.1 Motivation

TBD

14.2 US Aggregate Data

We want to look for some aggregate facts about the US economy and its trade patterns to get a sense of how some macroeconomic indicators had evolved around China’s WTO accession. In particular we will show that

• Expansion of Chinese Trade. Essentially all of US trade growth since the 1990s is from the expansion of Chinese imports.
• Fall in Real Interest Rates Around the time the Chinese trade expanded.
• Expansion of the Trade Deficit during this time period.

14.2.1 Load Necessary Packages

Install the ones you do not have yet.

library("fredr")
library("purrr")
library("dplyr")
library("readr")
library("tidyr")
library("ggplot2")
library("magrittr")
library("lubridate")
library("PerformanceAnalytics")

14.2.2 Get Data from Federal Reserve

We need the following variables from FRED:

codes = c("GDP", "IMP0015", "IMPCH", "EXP0015", "GS1", "CPILFESL")

We also need to tell FRED our API key to authenticate ourselves. For this course, you can use the following command:

api_key = "YOUR KEY"
fredr_set_key(api_key)

1. Use the fredr_series_observations command on a single variable to get the data for that variable
2. Now, find the option to pull only data starting at 1990-01-01. Make sure you format this number as a date.
3. What happens if you use the previous command on codes instead of a single variable name?
4. Solve it by applying the correct map function which returns a dataframe by row-binding. Save your dataset as df_raw

14.2.3 Data Transformations

For the rest of this part, take df_raw and save the transformed output as df:

5. Now split your data into columns
6. Rename your newly created columns as such:
   • gdp <- GDP,
   • imp_ch <- IMPCH,
   • imp_all <- IMP0015,
   • exp_all <- EXP0015,
   • t_bill <- GS1,
   • cpi <- CPILFESL
7. cpi is coded in billions of USD while exports and imports are in millions, multiply cpi by 1000 to have all values in millions
8. create additional variables for our date using the lubridate module. We want columns year, quarter, month, day that only contain this part from the date column. Find the correct functions in lubridate to achieve this
9. Sort your data by date

We see that gdp is coded quaterly while the imports and exports are per month. We need the data grouped annually and quarterly respectively for the next two parts:

14.2.4 Data Grouping

10. Group df quarterly into a dataframe called df_quarter. In this dataset you want

• start_date as the minimum of date,
• gdp, imp_all, imp_ch, exp_all all aggregated as sums (how do you deal with NA’s?)
• cpi, t_bill aggregated as averages

11. Group df annually into a dataframe called `df_year. In this dataset you want

• gdp, imp_all, imp_ch aggregated as sums (how do you deal with NA’s?)

14.2.5 Fact 1: Increase of Imports from China and the Rest of the World

For this exercise, we are going to use df_year

1. drop data from 2020
2. We want to create the following three variables:

• global_share = 100 * imp_all / gdp
• china_share = 100 * imp_ch / gdp
• nonchina_share = global_share - china_share

3. Create the following graph

• years between 1991 and 2008
• x-axis: years
• y-axis: china_share and nonchina_share (add to lines to your plot with different colors)
• a vertical line for x == 2008

14.2.6 Fact 2: Increasing trade-deficit of the US

For this we will work with df_year again.

1. drop data from 2020
2. We want to create a variable trade_deficit (imports - exports)
3. And a variable trade_deficit_share for the share of the trade deficit compared to the gdp
4. Plot trade_deficit_share over time
5. Pick a nice color
6. Add a vertical line to the plot for year == 2008

14.2.7 Fact 3: 400 basis point fall in real Interest rates leading into China Expansion

For this we will work with df_quarter again.

1. We need to calculate the inflation rate from the consumer price index (cpi) column
   • find out how to calculate this
   • look at the first lines of your dataset. What’s the problem?
   • ungroup the dataset first before you repeat the previous step.
   • look at the first lines of your dataset again.
   • you now have the quaterly inflation rate
2. Try to apply the Return.annualized function from the PerformanceAnalytics package to the newly created infl column inside mutate
   • what’s the problem?
3. We will solve this problem in two ways
   1. Use the function inside map_dbl instead (not in a mutate)
      • assing the output to a new column of df_quarter called annum_return
   2. Use Vectorize(Return.annualized) inside mutate instead an save the output to annum_return2
      • What does the function do?
4. Create a new variable called real_r as the difference between t_bill and one of your annum_return columns
5. Plot real_r against date
6. Add a vetical line at the date 2002-01-01
   • Hint: You have to parse the date first as a date and then convert it into a numerical value

14.3 Descriptive Stats in ADH using Weighted Statistics

We are now going to work the the ‘micro’ data directly from ADH. Luckily, some of our coding friends at NYU Stern have done a tonne of the heavy lifting for us and merged all of ADH’s essential data together into one file.

The trickiest thing to understand is the timing of the data and the variable names.

Here is some info:

14.3.1 Details about timing is as follows.

The start of the period is 1991 and then end is 2007. This is then divided into two periods. The first periods is 1991-2000, thus this is a 9 year time period. They convert stuff into a “comparable decadal scale” see Footnote 22. Thus, for values for this period, they multiply them by 10/9. The same issue arises for the second period which is 2000-2007. The values for this are again converted to “decadal scales” so they are multiplied by 10/7.

The Appendix Table 2, reports the income variable and the decadal adjustments. In the summary statistics for the stuff that we care about, the ADH data is adjusted in this way described above.

That is, variables starting with ‘l’ are in levels whereas variables starting with ‘d’ are the decadal equivalents.

As necessary, we will tell you which variable to use, so that \(*_somevariable\) means to choose the appropriate level or decadal equivalent. We leave you to figure out which of the \(l\) or \(d\) variables to use. Do ask us if you are confused.

14.3.2 Understanding the Essence of the Paper and What Comes Next.

Read Section 1 of ADH, so that you build an understanding of there main measure ‘IPW’ and what the paper is about. This will help you understand the context behind the remaining exercises in this notebook and those to follow.

Your first task will be to compute some descriptive statistics from the data. To be more precise, you will replicate some of the key numbers in Appendix Table 2 of ADH. (On a side note, at least one of us thinks this table should be in the main text!)

14.3.3 Load Necessary Packages

Install the ones you do not have yet.

library("readr")
library("tibble")
library("dplyr")
library("Hmisc")

14.3.4 Load Data

Like always, we are going to load the data and save it as a tibble

df = read_csv("data/adh_data.csv") %>% as_tibble()

14.3.5 Compute Simple Grouped Mean

1. Find which years (yr) are reflected in the data.
2. Compute the average number of chinese imports per worker ('l_tradeusch_pw & d_tradeusch_pw) for each “year”.

14.3.6 Computed Weighted Group Means and Standard Deviations

For the rest of the exercise, weight the mean by population count per region instead (l_popcount) and compare it with the numbers in the table.

3. Repeat step 2 with weights.
4. Now also compute the weighted standard deviations for both variables. Hint: Use the Hmisc package and find the relevant function.
5. Now compute the mean and standard deviation of the average household wage and salary (l_avg_hhincwage_pc_pw, d_avg_hhincwage_pc_pw)
6. And once more for share not in labor force (l_sh_nilf, d_sh_nilf)

How well do your numbers line up with those reported in the paper?

14.4 Regression Analysis

Let’s first load the necessary packages to read data and do fancy regressions:

library("readr")
library("tibble")
library("sandwich")
library("lmtest")

And let’s load the data like we always do:

df = read_csv("data/adh_data.csv")

14.4.1 OLS regression

The core of the paper is looking at what happened to laborer’s when theres an increase in us imports from china. Let’s try and replicate part of Table 9 - namely the estimate from panel A column 2.

Their y variable is relchg_avg_hhincwage_pc_pw. The important x variable is decadal trade between the us and china d_tradeusch_pw.

1. Run that simple regression
2. Now add heteroskedasticity robust standard (HC1). Hint: Use the sandwich and lmtest packages

Now we will start to add extra x variables.

3. Start by adding t2 - a dummy variable for whether observation is in the second decade. Fit again with HC1 robust standard errors.

14.4.2 Clustering

Let us now use clustertered standard errors instead. ADH cluster by statefip. Hint: use the felm package.

1. Run the basic regression with clustering
2. Add the following controls to your last regression:
   • l_shind_manuf_cbp
   • l_sh_popedu_c
   • l_sh_popfborn
   • l_sh_empl_f
   • l_sh_routine33
   • l_task_outsource
3. Add region fixed effects to your regression.
   • First find all variables in the dataset that start with reg_
   • Add these to your last regression

14.4.3 Instrument Variables

1. Instrument d_tradeusch_pw with d_tradeotch_pw_lag in your last regression
2. Weight your regression by timepwt48