I'm a retired software engineer who's been trading stocks/bonds/ETFs/futures contracts for many years. In recent years, I've been putting my math/programming skills to use developing my own algorithmic trading strategy. I started with the ideas in Ernie Chan's "Algorthimic Trading" book, in particular a cointegrated mean-reversion algorithm using a Kalman filter, and I've applied it to a cointegrated ETF triplet. Backtesting using 5 years of historical data suggests a 50% AYR in-sample, about 45% out-of-sample, with a maximum drawdown of 8.9%.
I've recently added a cointegrated VAR (Vector Auto-Regressive) model to the mean-reversion algorithm and improved the performance in backtesting. Trading with the algorithm live over the past two months, I've made a 7% return (which implies a 50% annual return if the algorithm continues to perform as well). As you tell, my approach relies pretty heavily on math and statistics, so it might not be for everyone. I prefer to do the algorithm coding myself rather than rely on "canned" trading programs, or "black box" approaches (such as neural networks) in which the trading rules are mysterious (and possibly spurious).
Anyway, if anyone else here is doing algorithmic trading, I'd be interested in hearing about your experiences.