Lately I’ve been asking Claude to replicate papers and test if common technical analysis beliefs hold up. Claude compiled the results and discussions into papers themselves, and they are up for grabs at newhighs.nl. Spoiler: almost nothing passes the regular statistical tests. A negative result is a result nonetheless, and it can save a lot of money by knowing you have to stay away. There’s quite a bit of statistics in the papers. Claude also made a guide for that.
One really amazing result was the discussion of Andrew Lo’s paper from the 90’s. In it he demonstrates that common patterns like head-and-shoulders, pennants, triangles and what else, do indeed carry information: the behavior of the market after such a pattern occurs is different from everyday markets. He did not say you could make money from it. Claude replicated the findings but came with a surprising explanation: the information Lo talked about was that after a pattern there is less volatility than on average. That is totally normal, because volatility clusters: calm days are followed by calm days, wild days by wild days. A pattern is by definition a very orderly period, so it is obvious that the days after a pattern will be calmer than average.