Thanksgiving Data Work

Setup

• The primary data are annual import volumes, in terms of value and weight, from China and Pacific Rim countries, to a select number of US ports, over 2013-2018, and at the HS2 (Chapter Level). I computed the mean value and weight of imports over the sample period by US port \(i\). Port \(i\) was selected if was in either the top 20 highest mean value or weight series. Additionally, I merged the Port of New York and the Port of Newark’s import volumes into a single composite port “NYC” as both ports operate under the Port Authority of New York and New Jersey.

• The top 20 ports were sorted into coastal regions: “East”, “West”, and “Gulf.” Note that Great Lakes ports were excluded.

• HS2 chapter data was aggregated according to the broader chapter headings (\(k\) hereafter) outlined here : https://www.foreign-trade.com/reference/hscode.htm.

• I computed port \(i\)’s share of total Chinese/Pacific Rim imports (value/weight) at time \(t\) per \(k\) (and for all \(k\)).

• The Panama Canal expansion affected US ports starting in 2016. This year is marked in all subsequent plots with a dashed line.

Total Port Shares

Total Port Volumes

The next series of plots depict aggregated trade shares in terms of volume/weight and trading partner by US port overtime.

Port Shares

Difference in Difference

I group East and Gulf Coast ports into a new category of “Non-West” ports. In context, \(Coast_i=1\) if port \(i\) is a “Western” port, 0 otherwise.

Regression Equation:

\[ Y_{it}=\beta_0+\beta_1 1\{t=2016\}+\beta_2Coast_i+\beta_3\Big(1\{t=2016\}\times Coast_i\Big)+\varepsilon_{it} \]

• \(Y_{it}\): includes import levels/shares from China/Pacific Rim countries at either the aggregate or \(k\) product type. Note that for some regressions, \(Y_{it}\) is log-transformed.
• \(\beta_1\): reflects the year that East coast ports could begin to receive Post-Panamax ships.
• \(\beta_2\): indicates the coast of port \(i\). In principle, this should be the treated \(\beta\); the treated group of ports is a subset of “Non-West” ports (e.g. NYC, Savannah, Charleston, and Baltimore).
• However, since we only have a total of 3 years of post-treatment for a small number of ports (both treated and untreated), its hard to get anything statistically significant for \(\beta_3.\)
• \(\beta_3\) : captures the interaction/dif-in-dif coefficient and in the following tables is given by V2.

No dif-dif estimation used non-log transformed data (either levels or shares for both aggregated data/product subseting) produced significant results. Consquently, no untransformed level/share dif-dif models are presented here.

Dif-Dif Logged-Levels: Aggregated Goods

The dif-in-dif here used log transformed levels. *For interpretation purposes, negative and signficant coefficients translate to “trade in West coast ports on average fell relative to non-West coast ports.”

	Country	Series	V2	Estimate	p.value	p_type
4	China	Value	Coast_2West:Year1	-0.2131801	0.0856690	<0.1
8	China	Weight	Coast_2West:Year1	-0.2673518	0.0398882	<0.05
12	Pacific Rim Countries	Value	Coast_2West:Year1	-0.1829167	0.1383096	>0.1
16	Pacific Rim Countries	Weight	Coast_2West:Year1	-0.2332781	0.0744484	<0.1

Just looking at East/West Coast ports (e.g. Gulf coast ports were dropped and so now \(Coast_i=1\) if port \(i\) is a “Western” port, 0 for Eastern ports), nothing is significant.

	Country	Series	V2	Estimate	p.value	p_type
4	China	Value	coastWest:Year1	-0.1587075	0.2189033	>0.1
8	China	Weight	coastWest:Year1	-0.1925998	0.1507367	>0.1
12	Pacific Rim Countries	Value	coastWest:Year1	-0.1485015	0.2429343	>0.1
16	Pacific Rim Countries	Weight	coastWest:Year1	-0.1915092	0.1492203	>0.1

Dif-Dif Shares (Logs): Aggregated Goods

The first regression here uses the West/Non-West region classification. Unexpectedly, the only significant results are in the value series.

	Country	Series	V2	Estimate	p.value	p_type
4	China	Value	Coast_2West:Year1	-0.1812369	0.0056526	<0.01
8	China	Weight	Coast_2West:Year1	-0.0147580	0.7563242	>0.1
12	Pacific Rim Countries	Value	Coast_2West:Year1	-0.1439514	0.0044880	<0.01
16	Pacific Rim Countries	Weight	Coast_2West:Year1	0.0189441	0.6559134	>0.1

Just looking at East/West Coast ports (e.g. Gulf coast ports were dropped and so now \(Coast_i=1\) if port \(i\) is a “Western” port, 0 for Eastern ports), we have the same result as above: log value shares fall in West ports relative to Non-West ports.

	Country	Series	V2	Estimate	p.value	p_type
4	China	Value	coastWest:Year1	-0.1421167	0.0385570	<0.05
8	China	Weight	coastWest:Year1	0.0013100	0.9791748	>0.1
12	Pacific Rim Countries	Value	coastWest:Year1	-0.1481564	0.0050924	<0.01
16	Pacific Rim Countries	Weight	coastWest:Year1	-0.0319618	0.4805295	>0.1

HS2 Dif-in-Dif: Logged Levels

Log-levels by West/Non-western Ports are used here. I present \(\hat{\beta}_3\) estimates via barplots to better visually summarize results. Due to over-plotting along the \(k\) product axis, I’d suggest hovering particular bars to learn more/toggle bars on/off by significance.

Highlights:

-Machinery and Metals product imports to Non-West ports increased from China and the Pacific Rim, respectively, for the weight series.

-Machinery product imports to Non-West ports increased from China only in the value series.

HS2 Dif-in-Dif: Logged Shares

Log Shares by West/Non-western Ports (dependent variable is log \(k\) share)

Within the Value Series:

-Stone/Glass and Metal products increased in Non-West ports while Animal products increased in West ports for Pacific Rim imports only.

Within the Weight Series:

-Wood products increased in Non-West ports for Chinese imports only.

-Transportation and Metals increased in Non-West ports while Animal products increased in West ports for Pacific Rim imports only.

-Stone/Glass products increased in Non-West ports for both Chinese and Pacific Rim imports.

Next Steps?

-Use more granular HS6 (recall HS10 port-level data is publicly unavailable).

-Directly compare NYC, Savannah, Charleston, and Baltimore to LA/Long Beach and any other major West coast hubs.