Relating Strategy Turnover, Trading Frequency, and Transaction Costs

# How can we determine the minimum viable holding period given granular data and transaction costs?


# How can we determine the minimum viable holding period given granular data and transaction costs?












Given assumptions around transaction costs (e.g., slippage, fees), and potentially downsampled tick or bar data, how can we derive the minimum holding period required for a strategy to be viable — independent of the specific strategy logic (except for perhaps long-only, short-only, or both)?

This seems like it would be useful to help decide at what frequency it's even worth looking for predictability. However, I haven’t seen much written on this and suspect that this may not be how things are typically done/the order in which things are typically done in practice.

#### Example Scenario

Suppose I'm working with 1-minute OHLCV bars and construct a simple long-only strategy that:

- Enters a position at the current close if exiting at the next bar’s close is expected to be profitable (net of a flat percentage-of-execution-cost transaction fee).

- Exits if holding until the next close is predicted to reduce profit versus exiting at the current close (also net of a flat percentage-of-execution-cost transaction fee).

In practice, this strategy rarely trades, seemingly because the predicted return from one bar to the next is too small to overcome. Thus, even over long periods, it barely enters or exits.

#### Questions

- Is there a principled way to estimate the minimum expected holding period (or return horizon) required to overcome known transaction costs?

- Do practitioners actually use this approach to guide strategy design (e.g., determine viable signal frequency), or is this the wrong order of thinking? Do they instead test for predictability across multiple horizons and then incorporate transaction cost awareness later?

Thanks in advance for any insight or references. I’m trying to understand what the workflow of strategy development looks like at a high level.

## Answer by lehalle (score 2, accepted)

https://quant.stackexchange.com/a/83967

Your question suggests that for a given frequency, the transaction costs are independent from the strategy. It is unfortunately not the case.

Simply think about mean-reversion vs. momentum (i.e. cross-sectional trend following):

- mean reversion should be less costly since you buy while the price goes down before it goes up,

- and momentum requires you buy what is currently going up.

The latter should thus be more expensive to trade than the former.

It means you need to respect the natural frequency of your strategy. Measure the time $t_p$ for you strategy to change its sign (I know it is not obvious: you have more than one line), for sure you need to trade twice faster than that. If you strategy makes $\mu$ bp per month and $f={2\over t_p}=4$ month${}^{-1}$ (ie ${t_p\over 2}=$1 week), then your transaction costs should not be greater than ${\mu\over f}$ bp.

Professionals are studying the capacity of a strategy using a market impact model, that is non linear in the traded quantity (your view on transaction costs is linear, i.e. in bp). See Briere, M., C.-A. L, Nefedova, T. and Raboun, A., 2020. Stock market liquidity and the trading costs of asset pricing anomalies. In 12th Annual Hedge Fund Research Conference. In this paper authors adopt this approach to obtain tables like:

you will find more tables for more trading strategies like this one in the paper.

[EDIT] The speed/frequency of a strategy (answer to one of the comments). It is difficult to estimate the frequency or speed of a strategy. The worst idea would be to try to compute something like: time to go from long to short, for each position, and then average, because it would be fragile to choices of thresholds.

My recommendation (used by a lot of practitioners) is to rely on the turnover:

$$\mbox{Turnover} = {\mbox{Dollar volume daily traded (buy+sell)}\over\mbox{Average NAV}}.$$

This metric is clear: it is how much of your portfolio value to change every day. Hence

$${1\over \mbox{Turnover}} \mbox{ is a speed},$$

answering to how many days on average to change 100% of your portfolio.