New Millennium, New Market

The turn of the millennium was one of the craziest times in the history of stock markets. Of course, everyone knows about the "dot-com" companies with P/E ratios in the stratosphere. Fewer people understand how the "Y2K bug" forced so much spending on tech that it drove up valuations towards the end of the 90s (and the Fed helped by pumping money). But almost nobody talks about how the market changed. The best example of that change is this chart showing the 1-day autocorrelation of the S&P 500:

It's pretty easy to see the regime change there. After a half-century of positive autocorrelation, the market flipped to negative autocorrelation. Why did this change happen? Nobody knows, but it probably wasn't just one thing. I think it probably had something to do with shrinking trading costs, faster transmission of ideas, and increased reliance on automated trading systems. Regardless of why this change happened, I just want know how it affects trading strategies.

The "Buy the Rip" Era (1950 - 2000)

In this era of positive autocorrelation, the following trading algorithm would have made a killing.

• If SPX was up yesterday, hold SPX today
• Else, hold cash

The backtest results are impressive. This simulation doesn't take into account taxes and fees though. I don't think this type of strategy would have been feasible to execute back then due to high fees. Again, that's probably a big part of the market change.

The "Buy the Dip" Era (2000 - Present)

In the modern era of negative autocorrelation, the opposite strategy is preferable:

• If SPX was down yesterday, hold SPX today
• Else, hold cash

Again, this simulation doesn't take into account taxes and fees, but the outperformance is pretty significant. It's pretty neat to see how this change in the market flipped everything.

What this Means for Backtesting

This change poignantly demonstrates the biggest risk in backtesting strategies. The data can show an extremely strong pattern for decades, and then the pattern can flip on a dime. Of course, the digitization of markets around the turn of the millennium was possibly a once-in-a-lifetime change.

I generally disregard market data from before the 2000s. I don't really care how a strategy would have performed in the days where you'd learn everything from the morning newspaper, call your broker on a landline, and make an expensive transaction to purchase paper stock certificates. The market has changed. We live in a world of instantaneous news where transactions are cheap and robots do most of the trading.  Buffett and Munger are buying tech stocks. Long live the new regime!