Relationship : Quantitative strategy testing & historical data

[occulti]

Hello,

When testing a quantitative strategy, should we test it for the market we're going to use it for ?

Exemple :
I have 2 strategies, but I'm trading AAPL, S&P500 and Russel 2000 historical data.

My questions are :
1 - When we build / think of a strategy, is it just for one specific market or is it for different markets ?
2 - Market conditions are different, but we all know that markets are correlated, a profitable strategy could be profitable in a stock but not in another, or couldn't it ?
3 - If a tested strategy performed well (good Sharpe ratio, good returns curve, tolerable drawdowns) is supposed to be applied on real time, then why testing in the first place in historical data ? Why do we call it "testing" ? Doesn't "testing" mean reproducing all (or most of) common cases that the algorithm may execute ?

I am just confused with all of this. Could you please provide me with a simple, but a successive reasoning ?

Thank you all

 

[rajanS]

Hello,

When testing a quantitative strategy, should we test it for the market we're going to use it for ?

Exemple :
I have 2 strategies, but I'm trading AAPL, S&P500 and Russel 2000 historical data.

My questions are :
1 - When we build / think of a strategy, is it just for one specific market or is it for different markets ?
2 - Market conditions are different, but we all know that markets are correlated, a profitable strategy could be profitable in a stock but not in another, or couldn't it ?
3 - If a tested strategy performed well (good Sharpe ratio, good returns curve, tolerable drawdowns) is supposed to be applied on real time, then why testing in the first place in historical data ? Why do we call it "testing" ? Doesn't "testing" mean reproducing all (or most of) common cases that the algorithm may execute ?

I am just confused with all of this. Could you please provide me with a simple, but a successive reasoning ?

Thank you all

It would def be more robust if your strategy works in multi environments and across different time scales.

You are just backtesting with historical data. One way you could back test correctly is to split ur sample into two, use the first sample to estimate the parameters of ur strategy and apply that strategy to the 2nd time interval using the parameters obtained from the first interval. Testing is done to see whether ur not overfit!