Rol's Trading Journal
- Thread starter Rol
- Start date
I got burned today by CYH, Community Health Systems. I saw it heading rapidly lower from my entry but I saw no news out on it yet. Well, eventually the news came out that they were being sued by Tenet Healthcare over Medicare billing fraud, but by then it was down big. I am currently down 1K on 100 sh.
I will be adjusting my entry rules filter slightly so that todayâs entry would have been avoided in the first place. Also, I am going to try using a 10% stop loss on all of my system trades. The majority of my trades do not exceed 10% DD, and the ones that do, often seem to blow right through 10%. What I would like to achieve is a reduction in draw downs on news related stock moves, while at the same time, not interfering with the day to day gyrations in my overall portfolio.
I will be adjusting my entry rules filter slightly so that todayâs entry would have been avoided in the first place. Also, I am going to try using a 10% stop loss on all of my system trades. The majority of my trades do not exceed 10% DD, and the ones that do, often seem to blow right through 10%. What I would like to achieve is a reduction in draw downs on news related stock moves, while at the same time, not interfering with the day to day gyrations in my overall portfolio.
The data is from daily bars, O/H/L/C. In back testing, exits occur at the close of the day if the exit criteria are met. The entries occur if the entry price is between the H and L for the day. Liquidity filters are in place to attempt to eliminate spurious entry signals, but of course it is not 100%. I have limited symbols to those trading over 1 million shares per day, and it is still fairly robust. I plan to increase slippage on the exits though, since exits are at market. I figure that there is often enough volume around the close to help reduce slippage.
I cannot take every trade possible, but neither does the back tester. There are a finite number of available slots, and when they are filled, no more trades are taken by either me or the back test engine. Iâm not sure what serious issue could arise from this. I wish I could take every trade signal I got. When looking at past paper trades, I spend far more time wishing I could have taken the trade, than being thankful that I avoided taking the trade. When optimizing the number of positions allowed in back testing over several years, I did not see a tampering off of ending net equity until the maximum number of positions allowed exceeded 100.
I took my lumps today with CYH around the opening gap up. $800 loss. Currently underwater on most of my positions.
Real-time Unrealized P/L ($731.87)
Real-time Realized P/L (Today) ($413.08)
I face the same problem and I used to have the same approach. I always thought since the entries are randomized across the day that there's no issue with this - I was wrong.
I found that out by rewriting the code to work on intraday charts, the results gave me quite a scare. During the more serious downturns such as in 2002 and 2008, the results were dramatically different; in the more smaller drops, the results were worse but not as significantly.
When these black swan events occur, all components drop with high correlation and randomizing the entries would give you better entries than you would realistically get. I can guarantee that your results would be different for the second part of 2008, for the other periods - it depends on the trades taken in simulator/total possible trades ratio. You should also include the now defunct companies that had a lot of liquidity but took a nosedive that year, survivorship bias is a very relevant factor for these types of strategies.
I found that out by rewriting the code to work on intraday charts, the results gave me quite a scare. During the more serious downturns such as in 2002 and 2008, the results were dramatically different; in the more smaller drops, the results were worse but not as significantly.
When these black swan events occur, all components drop with high correlation and randomizing the entries would give you better entries than you would realistically get. I can guarantee that your results would be different for the second part of 2008, for the other periods - it depends on the trades taken in simulator/total possible trades ratio. You should also include the now defunct companies that had a lot of liquidity but took a nosedive that year, survivorship bias is a very relevant factor for these types of strategies.
Couldnt he just run multiple back tests, each taking random generated signals and compare them to each other, and if each backtest produces similar results he could say that its good enough.
I think his universe is sufficiently large enough that if he takes enough sample signals, his results would ultimately reflect the population. He keeps his size so small and number of positions large. But still good points you bring up, he should see if multiple random backtests are similar.
DGunz, that would certainly help. He could run a monte carlo simulation on the trades, 5% of the worst results would imitate real life performance roughly. The one problem I see is that during a 2008 type panic, Rol's system would buy some positions in simulation in such a way that they would get exited next day with profit and the system would initiate new positions, in reality he would be left with losers (because the positions were initiated in the first phase of panic) and maximum exposure, unable to pick up new positions.
It's much easier to pick up intraday data from some vendor and do the "true" backtest. In a bull market such as this, results will match simulation but the same won't happen in a volatile bear market.
Should he do that, I believe the results will still show high profitability but the equity curve won't be as perfect as it is now.
It's much easier to pick up intraday data from some vendor and do the "true" backtest. In a bull market such as this, results will match simulation but the same won't happen in a volatile bear market.
Should he do that, I believe the results will still show high profitability but the equity curve won't be as perfect as it is now.
oh I see what you are saying now...but hasnt Rol accounted for that. He stated that if his backtest is maxed out on a certain day his backest system will not enter in any more positions until he exits his losers. So in a way his backtest would also suffer through the first phase of the panic.
I totall agree with you about the ,monte carlo sim, he should just back test every possible signal, and then do a monte carlo sim, picking out a limited amount of signals because he has limited BP and cant trade it all.
Not exactly. When there is high correlation selling, the positions would be initiated during the first phase, for example, between 9:30 and 11:00. The second phase, from 14:00 to 16:00 would also trigger signals. Randomizing entries loses time priority. So if you could take 20 trades, you would already have full exposure before the second phase could start, you are essentially betting that the trades from the first sell-off outperform those from the second sell-off in the following days - but in my experience, this often is not the case.
Monte carlo is better than nothing but the only bulletproof way to know is with intraday data.
Monte carlo is better than nothing but the only bulletproof way to know is with intraday data.