When someone designs a mechanical trading system using closing prices trend following or other wise. People normally execute the same day and add slippage and costs, by this I mean take last 50 days of data points (including today) and execute the trade today.A) The problem with this scenario is that you never know the “real” closing price. So you must sample price lets say 30 to 45 mins before close, get your signal and to execute trades when there is reasonable liquidity. The problem is that sometimes the price when sampled can be very different from the actual closing price and may even show a 1% swing against you. The obvious solution is to add more slippage to counter that.B) Alternatively you can take today’s close to get the right signal and execute tomorrow at the open or near the close. Close seems a better place to do it as the system is using closing prices? Would you all agree with that? The problem of slippage still hasn’t been solved though, when best the execute. The benefit of executing the next day is you could get a lot of positive slippage but it could easily go the other way as well.The reason why I have contemplated this delayed scenario is that I am worried about the integrity/information capacity of the data point generated 30 mins before close. Alternatively this maybe the right price as this is a high volume period.PS: Liquidity analysis has shown that volume is highest in the first and last hour of trading, which essentially looks like an inverted bell shaped curvePPS: I have modelled both these scenarios and the performance goes down albeit by “a very small amount” when I execute the same day. What could possibly be the reason for this?PPPS: “a very small amount” can sometimes have a significant impact on the compunded returns right?

Any ideas?

I would always use the prices you can trade on to develop your trading system. You don't care about the 'real' closing price (which after all might have never been a traded price - depending on the market you are in).And, yes, this means the price can move for or against you. To do more stringent testing, you should tighten the rules:Base your system on quotes you can execute, specify the times you want to trade during your (back-)testing, for instance always trade at the executable level on 11am.Using close prices, slippage, and cost will usually give you a rough awakening...

QuoteOriginally posted by: ZedI would always use the prices you can trade on to develop your trading system. You don't care about the 'real' closing price (which after all might have never been a traded price - depending on the market you are in).And, yes, this means the price can move for or against you. To do more stringent testing, you should tighten the rules:Base your system on quotes you can execute, specify the times you want to trade during your (back-)testing, for instance always trade at the executable level on 11am.Using close prices, slippage, and cost will usually give you a rough awakening...All very good and valid points Zed,Unfortunately I can’t get execution price for 10 years of data (1993 to 2003)… I have access to some prices data, stamped hourly but only 2 years worth and very recent, which is not enough for me, although I can simulate some historical prices based on the current data but I’m not sure how good that may be as it will be perfect curve fit on the current vols.I haven’t had a rude awakening yet on my models (thank god).

I agree with Zed that you should model on tradeable prices. However you would still have to contend with market impact and slippage. Two important issues if you want to consider rampability. A lot depends on the execution strategy and the type of trading you are doing. For example if you are doing a market neutral basket your slippage may not be too high on an average. However if you are looking at trend following strategies then the slippage can be very crucial and might make or berak your strategy. That said a lot depends on your execution strategy. When I was trading european equities, the trading desk allowed me to do basket trades. I would send the basket out at 10:00 a.m. London time and the desk would give me a the price at that point at an average of 15 Bps cost. In such a scenario I used the historical 10:00 a.m. prices to test the model; when calculating the P&L I took into account the 15Bps trading cost. If you are currently trading, I would suggest that you look at the average price you are getting and then compare that with the previous day's close. Calculate the transaction cost for the whole basket with respect to the close price. When you do back tests using historical data assume that the model fires on the close price and when calculating the P&L include the transaction cost. Hope this helps.