Within the fixed income space, there's a lot of literature on PCA trading.
The first 2-3 principal component factors (PCs) can typically explain 90-99% of the total variances in yield curve movement. It's also nice, because the first PC looks like a change in the overall level of the yield curve, the second PC looks like a slope change, while the third factor corresponds to change in the overall curvature of the yield curve. So if you neutralize the first 2-3 PCs, you can indeed trade on the mean-reverting behavior of the residuals.
PCA is most commonly used for structuring so-called "butterfly trades." In this, you're neutralizing the first two PCs (level and slope) and trade on the third PC (curvature). For example, after running a PCA on 2y, 5y, and 10y yields, you may conclude that 5y yields are too high relative to 2- and 10-year yields (i.e., 5-year bonds are "cheap"). In this case, you'd buy 5-year bonds, while simultaneously shorting 2- and 10-year bonds. PCA comes into play, because for each unit of 5-year bonds, you have to choose appropriate units of 2- and 10-year bonds ("risk weights") so that the the first two principal components are neutralized, allowing you to trade any abnormalities in the third principle component.
The best literature I've come across is Salmon's Principles of Principal
Components, which is easily available by Googling. It also includes extensive backtesting results from butterfly trading. Another (not as good but still pretty good) one is Morgan Stanley's "PCA for Interest Rates."
I would point out that trading on PCA mechanically is usually not a great idea. A lot of macro-factors can disrupt "well established" patterns in truly splendid ways. After the financial crisis, 5- and 7-year bonds looked rich based on many statistical methods, PCA included, but they just kept getting richer...