9
$\begingroup$

I am working on an intraday strategy using 5/10 minute bars. I am getting a decent return and sharpe on the strategy. But on close examination I see that I am making about 1 cent per trade (I haven't taken transaction cost into account).

How do you guys go about choosing a good intraday strategy ? I have always used sharpes and CGAR for long-hold time strategy but seems like evaluating intraday strategies is a different beast.

I also notice that my strategy makes money for symbols in certain sector, and not for other ( this is my backtesting dataset). Do you normally throw out symbols with negative returns when doing a forward test as part of your portfolio ?

EDITED: This strategy is based on US equities and yes I am using a combination of midprice and previously traded price. Question is quite simple I think, how do you guys evaluate intraday strategies ? How big of a factor is slippage intraday ( on a 10 minute scales for example). ?

$\endgroup$
6
  • $\begingroup$ Can you describe a bit about the strategy? Intraday stuff seems to depend more on statistical anomalies and price patterns. Longer term strategies might take sector bets with a factor model. I'm a bit confused why certain sectors might behave better than others in a short term strategy. $\endgroup$
    – strimp099
    Commented Nov 23, 2011 at 17:09
  • $\begingroup$ To echo strimp099: can you tell at least is it US equities, futures, forex? Do you trade at midprice? $\endgroup$
    – LazyCat
    Commented Nov 23, 2011 at 18:57
  • $\begingroup$ This strategy is based on US equities and yes I am using a combination of midprice and previously traded price. Question is quite simple I think, how do you guys evaluate intraday strategies ? How big of a factor is slippage intraday ( on a 10 minute scales lets say). ? $\endgroup$
    – silencer
    Commented Nov 23, 2011 at 23:58
  • $\begingroup$ if you are making 1 cent per trade, you are effectively scalping, no amount of market data or simulation would be able to determine your fill ratios. a pessimistic way to backtest would be eg: if you are buying at $1 you will only be considered as filled if the traded price goes below $1. $\endgroup$ Commented Nov 24, 2011 at 9:49
  • $\begingroup$ Many questions in the same thread, why don't you split this? $\endgroup$ Commented Jan 5, 2012 at 3:27

2 Answers 2

2
$\begingroup$

If you are making 1 cent per trade before slippage and commissions, I don't necessarily think this is a viable strategy - commissions alone might evaporate that edge. If you are adding liquidity (as defined by the exchanges) perhaps you can obtain rebates to offset commissions and slippage.

Slippage can vary depending on the securities being traded. Further, it includes eating the spread and moving the market based on your size. For an intraday strategy, I assume you'll be trading the most liquid securities you can find that will reduce the spread slippage. When backtesting, use different size spreads for slippage estimation. For example, 1 cents, 2 cents, 3 cents, etc. See what your Sharpe ratio looks like at each level of spread so you can get an idea of how the strategy might perform in times of market stress (illiquidity).

In terms of getting worse prices for your trades because of size, that really depends on your strategy. If you're trading one or two round lots of SPY, chances are you can get filled without moving the price. That's a bit harder to determine.

In general, you can simulate your strategy with different parameters and assumptions and see what combination of parameters generate the highest Sharpe ratio. Build your backtest so it takes a variable input for slippage. As the slippage increases, your Sharpe ratio should decrease. When the Sharpe ratio gets to a certain unacceptable level, you should know what slippage you need to do better than.

$\endgroup$
1
$\begingroup$

Your PnL is a function of traded volume, transaction costs and mills p/share (mills meaning $0.001 p/share). You also have to include exchange rebates and fees, a significant adjustment given your backtested results.

1 cent net p/share is actually a good return for high-frequency or market making strategies that trades in the millions shares p/day. However, given the time periods you indicate I doubt your strategy is trading that kind of volumes. Then there is adverse selection which of course will lower your net PnL. Finally you have to consider SEC taxes and your broker or proprietary infrastructures cost. Market impact (slippage) is a function of your order size, stocks liquidity and other factors. There are Phd treatises on the subject, it's impossible to give a valid answer in two sentences.

In any case Sharpe and profit p/share are good metrics for evaluation. That is what I use during model development.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.