Frosty's trading bot goes live part 2

[walterjennings]

Quote from GTS:

I guess we are talking about two different things - your proof seems to refer to max (unrealized) drawdown of a trade while the quote that I cited and refuted is referring to the realized loss or profit of a trade:
Since that statement said "loss" not "drawdown" I stand by my assertion that it is false.

Adding a stop can most definitely result in a greater loss on a specific trade.

i guess i didnt make my statement clear then. when i said
"surprising. by drawdown you mean daily total right? a stop added to your current system should not result in any greater loss on a specific trade." i meant to be asking frost wether the increases were seen as an increase in avg drawdowns (> 1 trade) or an increase in drawdowns caused by a single trade, my error was using the were 'specific' instead of single.

yes your statement is correct. a stop can result in a larger loss on a trade. much the same as a stop can result in a smaller loss on a trade, we cant predict the future movement of the market after any decision. in other news fire hot and scissors sharp.

 

[dcraig]

Quote from GTS:

Its telling that folks that actually have done systematic backtesting have one position regarding stops while it seems those in the other camp advoacting stops do not seem to have any real experience in this area.

Its definitely not intuitite - I must admit I was surprised the first time I applied a stop to strategy I had and saw how much worse the results became. Then I analyzed it, trade by trade and saw what is happening.

Its very rare for any trade to be immediately profitable upon entry. The majority of trades spend some amount of time underwater. If your stop is too tight then you turn many winning trades into losing trades. If your stop is too wide then it really doesnt help because you still take heavy losses on those stopped out trades and you still miss out on a few trades that would have turned around and ended up positive.

In a trend following system you really shouldnt need a stop. Assuming that the system trade exit criteria is watching that the trend has reversed then trades should naturally exit on their own anyway (if you are long and the trend turns down and stays down then the system should want to exit on its own).

Quite right - it is not intuitive and the more system testing one does, the more surprises one encounters and the more a lot of conventional "wisdom" turns out to be worse than useless as does frequently ones "intuition".

I'm not advocating stops or no stops. I think you have to do the testing and work it out for yourself.

But there are many ways to control risk and stops are just one of them. Being market neutral or at least partly market neutral is another. And diversification another. Thinking about other alternatives is important for a systems trader.

 

[maxpi]

Quote from dcraig:

Hmmm ...

I have been experimenting with a reversion to mean type system applied to a (screened) basket of stocks eg DOW 30. It trades intraday - always flat at EOD.

I have not been able to find any stop loss strategy that improves any measure of performance - net profit, maximum drawdown, profit factor, Sharpe ratio etc. You name it - they are all worse if stops are applied. The simple measure of closing all open positions with MOC order beats anything else for positions that do not hit their target during the day.

As I said in a previous post, diversification can be the best way of dealing with unacceptable risk.

Not to nitpick but MOC is a time stop. This is not the first example I have heard of from a system developer that time stops were best for what he was doing.

 

[walterjennings]

frost, it would be interesting to work out mathematically what the max distance the price can move against (worst case price action) before the change in the trends cause your system to go flat, basically figuring out the limit of your dynamic stop setup. have you already looked into that? from the recent results it seems it will be > 1000$ movement

 

[dcraig]

Quote from maxpi:

Not to nitpick but MOC is a time stop. This is not the first example I have heard of from a system developer that time stops were best for what he was doing.

Agreed, it can be called a time stop. But this is fundamentally different from stops that are derived in some way from entry price or the instrument price time series. Perhaps this says something about price time series forecasting ?

I'll also reiterate that is a reversion to mean type system, so I'm claiming nothing other than it is difficult to find effective stops for this system. The clue, I think. is in the words "mean reversion" as to why this may be so.

 

[RichardTodd]

Quote from GTS:

Its telling that folks that actually have done systematic backtesting have one position regarding stops while it seems those in the other camp advoacting stops do not seem to have any real experience in this area.

I guess a person's choice really depends on their goal for the system. To maximize expected performance, all my tests agree with everyone else's tests that stops just get in the way.

But, there are reasons a person might want stops, anyway. I can do tons of backtesting and verify that a coin toss comes up tails about 50% of the time. But, that doesn't mean that I won't hit a stretch in the future where it comes up tails 1000 times in a row. or 100,000 times in a row, for that matter.

So, if a goal is to minimize risk of ruin, then you might keep your eye on the probability of x consecutive losses. Then, you might add things like variable or hedged position sizing, scaling in/out, and stops, to tune your system to bring the risk of ruin to a level that you can live with. You'd do this even though it often hurts backtested historical performance.

I agree with a previous poster that diversification would work well to reduce risk without stops, as long as you have multiple non-correlated systems. Unfortunately, I don't! So, I do what I can...

 

[BlindLemonBoosh]

Quote from RichardTodd:


I agree with a previous poster that diversification would work well to reduce risk without stops, as long as you have multiple non-correlated systems. Unfortunately, I don't! So, I do what I can...

Are you able to run the same system on non-correlated indices, or something to that effect?

 

[frostengine]

My second positive week in a row. Still way down, but hey at least its positive weeks. I don't mind a slow grind back up as long as its heading upward....


Today +$215.20
Week to Date: +$28.2
Since going live: -$2885
Total days: 17
Avg per day: -$169.70

Up Days: 9
Down Days: 8

 

[nkhoi]

Quote from frostengine:

My second positive week in a row. ..

just what I suspect, your bot is also capable of handle trend day as long it is not a fast moving market. You get about 20 days like that out of a year, now if you put in a routine for fast market then back test it and the routine fires about 20 signals a year then you are all set.

 

[trader225]

This is really a great thread. If anyone thinks that I have been disruptive by railing against "bar trading," use of no stops, etc., I hope that they'll pardon me. I am trying to learn about bots and may react inappropiately when the adrenalin flows. For example, with the ES I have seen ema crossover charts (ema4() and ema(8) on 5 min. charts) that lead me to believe that some money could be made with a crossover bot, but everyone seems to say something like, "nooooooo, go back to school." Well, I don't do backtesting. I do realize, however, that a decision must be made at some arbitrary point, whether it's time base or tick based. It just seems to me that something based on that sort of ema crossover, for entry, not exit, could make some money. I do know, however, that I don't know how to codify the exit.

p.s. I have been looking at IB's charts, for whatever they're worth.