Choosing Data Frequency for a Short Backtest

Summary

The document considers whether daily or weekly observations are adequate for testing a strategy over a six-month lookback. It argues that weekly data would yield too few observations for meaningful estimation and suggests intraday data as one way to increase the sample. Alternatively, it proposes lengthening the strategy horizon so that the test can use a broader period of historical data.

The answer’s reasoning is that a larger sample can improve the accuracy and precision of estimates and reduce sampling noise. It does not provide calculations, empirical comparisons, or details about the strategy, market, or data quality. More observations alone do not guarantee a reliable backtest: dependence between observations, transaction costs, selection bias, and whether intraday data matches the strategy’s actual trading horizon can all matter. The advice should therefore be treated as a general sampling consideration, not a complete method for choosing data frequency.

Key ideas

• Data frequency affects how many observations are available for a fixed testing period.
• The answer considers weekly observations inadequate for estimating a short-lookback strategy.
• Intraday data or a longer strategy horizon are offered as ways to increase the usable sample.
• A larger sample may improve estimate precision, but sample count alone does not establish backtest reliability.
• The document gives no empirical comparison or strategy-specific guidance on frequency.

Tags

# Daily or weekly data?


# Daily or weekly data?












I am trying to test a strategy but do not know if I should use daily or weekly data? I have tried researching this & have found that daily data will bring " too much " noise compared to lower frequencies. Specifically, I am trying to "backtest" using a 6-month look back period. If I use daily data I will have approximately 125 observations versus 26 if I use weekly. Will I suffer from "too much" noise or lack thereof if I use weekly?

## Answer by Quantopik (score 1, accepted)

https://quant.stackexchange.com/a/16564

If you need for backtesting a trading strategy for a 6-month look back period you will have to use intraday data, in order to be able to get more significative estimates.

26 observations are not enough to implement a backetesting model and, generally, larger and larger number of observations leads to increase the accuracy and precision of your estimates. Look here for a simple explanation of that.

Secondly, increasing the number of observations reduces the sample noise and not increases that ones because of the same reasons explained in that link I posted above.

A solution could be use intraday data to increase the sample or setting your trading strategy on more long term.

Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.