I’m giving a presentation on what determines Wisconsin exports. Here’s an interesting finding – since 2018, real goods exports have underperformed what would’ve been expected based on rest-of-world GDP and the real value of the US dollar.

Here’s Wisconsin goods exports (from BEA) deflated into real terms using the BLS export price index for all commodities, plotted with the index of real rest-of-world trade-weighted GDP.

**Figure 1:** Goods exports of Wisconsin in bn 2000$, SAAR (blue, left log scale), and real rest-of-world GDP, 2005=100 (brown, right log scale). NBER defined peak-to-trough recession dates shaded gray. Source: BEA/Census, BLS both via FRED, Dallas Fed DGEI, NBER, and author’s calculations.

I use an error correction model involving exports, rest-of-world GDP, and the real exchange rate, estimated 1996-2017. The long run elasticity of Wisconsin goods exports to rest-of-world GDP is about 1.1, the elasticity with respect to the exchange rate of about 0.40. I use this equation to dynamically forecast 2018Q1-onward, and obtain this (note: units are in millions, monthly rate):

**Figure 2:** Goods exports of Wisconsin in mn 2000$, seasonally adjusted (black, log scale), and dynamic forecast from ECM (blue), +/- 1 standard error confidence interval (gray lines). NBER defined peak-to-trough recession dates shaded gray. Source: BEA/Census, BLS both via FRED, NBER, and author’s calculations.

The increasingly large forecast errors are associated with the beginning of the trade war. In other words, the historical correlations obtaining up to 2017 implied continued — albeit slow — growth in real exports through 2019, consistent with growth in the world economy, and a strong dollar.

Quantitative importance? Over the two years of the trade war preceding the onset of covid-19, the forecast errors average about $1 billion (in 2000$). With 2021 real exports at about 14.6 billion 2000$, that’s on the order of a 0.7% reduction.

The forecast error could be attributed to model misspecification, or the influence of other important determinants that were omitted from this three variable specification (the adjusted R^{2} is about 36%). It’s always hazardous to argue on the basis of residuals, but at first glance, given what we know about retaliation levied by our trading partners against swing state, the trade war and retaliatory actions seems not an implausible source of the forecast error.

So (once again), thanks Trump!