The paper alternatives between using eigenportfolios and sector/industry ETFs for statistical arbitrage. For instance, sections 2.1-2 vs. 2.3.
The trade in Section 4.1 is long some stock and short an appropriate amount of sector/industry ETFs.
That being said Sections 5.3 and 5.4 discuss PCA strategies in a backtest, with relatively little additional information. It seems they are using the approach described at the beginning of Section 5 (basically they regress the stock against 15 or so eigenportfolios and go long the stock and short the eigenportfolios based on the beta, assuming I'm reading it correctly).