I agree with the article. I've been arguing on other threads that technical analysis doesn't work very well, for the simple reason that market prices are mostly a random walk. This means that analyzing a stock chart offers very little ability to predict future prices.
Autocorrelation is the cross-correlation of a signal with itself at different points in time. A positive autocorrelation enables one to use momentum strategies profitably. As the article shows, there was a fair amount of positive autocorrelation in the stock market until about the mid-1980s. Since 1986, the market has periods of both positive and negative lag-1 autocorrelation, but mostly negative. This makes it very difficult to implement momentum strategies.
Autocorrelation is related to the autoregressive model, which says that a market price depends on the most recent price (or prices), plus a random term. A common test to determine whether a time series is a random walk is the Unit Root Test. When I apply the Unit Root Test to 40 years of daily S&P 500 prices, I get a p-value of 0.978 which suggests that a random walk is highly likely. (A value of 1 would be an exact random walk.) A random walk means that price changes are a purely random variable that cannot be predicted from past prices.
I favor mean reversion strategies, but applied to a cointegrated portfolio, NOT to single asset prices. A cointegrated portfolio is one in which the individual assets have been carefully selected and weighted so as to reduce the random component and increase the deterministic (predictable) component. I have had success trading such a strategy.
In spite of all this, there are some on here who insist that market prices are not random, and that technical analysis can work. I wish them luck, but I'm not willing to ignore the statistical analysis which says that prices are very nearly a random walk. Some studies have shown that the stock market is about 95% random, and 5% deterministic. Those are not good enough odds for me.
