%%I understand points 1 and 2. I fail to understand point 3. Why should longer time-frame work better than shorter time-frame in backtesting? Shouldn't shorter time-frame work better since you have more data for testing when you go into intra-day?
%%
Because small samples tend to be more in-acurate/not accurate. Small samples tend to not even catch a bear trend or bull trend or sideways slop chop.
Sure could learn something on any time frame in 20 years; usually more=20 years shorter term data]= less profits.Market makers/specialist are one exception to rule, usually. Good question........................................................................................................
%% That sounds right, JBaX;I guess his question is more related to the resolution of your data (as in 1 minute candle history for the last 2 years instead of 1 day candle history for the last 2 years).
In that case the better the resolution of your data the better your model, and in my opinion the shorter the time frame, the less random your data will be: more reliable patterns can be found in the smaller time scale - these patterns are result of algorithmic rule-based trading.
In larger time scales the market activity is a result of many random interactions (news, legal rulings, sentiment, social hype, business decisions, etc) so you will be essentially guessing.