Sina made this video warning against overfitting. The general advice is good but he is wrong on this particular example. This is an important topic given many people follow this supposed delay between GL and the BTC price. Initially I thought that the delay idea was bogus. But I always double and triple check everything before I arrive to a conclusion. After many tests I concluded that indeed GL is predictive of Bitcoin price movements under the assumption that the prediction is directional (up or down). The best delay for this is 85 days. It is true that is pretty silly to claim every twist and turn in GL corresponds to a price action in Bitcoin. On that I agree. But in general there is predictive power in GL. Data says so. If you know what to look for. In fact, GL can predict about 72 % of the time a movement up in price if we smooth the price over a week and we focus only on directional moves. So it is important both to be rigorous but also creative in data analysis.