Weekly results of a wanna B algo trader - 2011

[jedwards]

Quote from DGunz:

Jed,

Seems like most people who trade mechanically successfully, trade as much as their strategy can bear, or their risk tolerance can bear. If you have winning strat you should always trade. The casino never stops for the night just because they are up, they only limit the size of the bets to insure they can out live the down draws. If you are considering stoping when you are up by a lot, that can either mean you are trading too large size or your strategy does not have a strong enough statistical edge for you to be confident to keep trading through it.

fwiw, I always paper trade new strats until I find a small edge. Sometimes it never even makes it live...There is no shame in paper trading until your are certain your live trading will make money, that way you wont have hesitation.

Yes, I'm well aware of the "casino vs gambler" perspective I've been struggling with this particular problem myself, because I force myself to trade in order to try to be the casino as opposed to the gambler.

But just like automated traders look to find an edge for the behavior of a setup on an instrument, can't we use the same statistical analysis on the PnL curve of our own setup? It would be similar to taking the second derivative of the trade and doing an analysis on that, no?

Charting the PnL curve of a setup and applying meta-rules on top of the setup seems to also be a legitimate way to trade.

For example, some example rules could be if you get 4 wins in a row, you can stop trading the rest of the week. Or if you get 4 losses in a row, double the number of contracts. Things like that can also be backtested to see how they affect the profitability of the setup overall. They're not trading setups per se, but they seem like legitimate modifications to a trading strategy, no?

 

[DGunz]

Quote from jedwards:


Charting the PnL curve of a setup and applying meta-rules on top of the setup seems to also be a legitimate way to trade.

For example, some example rules could be if you get 4 wins in a row, you can stop trading the rest of the week. Or if you get 4 losses in a row, double the number of contracts. Things like that can also be backtested to see how they affect the profitability of the setup overall.

I utterly and complete understand where you are coming from. I think if you did back test you might find is this: a winning strategy will win less, and a losing strategy will lose less. Assuming you have a winning strategy then its going to hurt you.

You bring up the point of consecutive wins/loss days, and using pnl as an derivative of strategy optimization, and its really tempting to do so, I will admit, but I suspect that the cold hard facts are that the system's expectancy does not change based on what has already happened. The tenth consecutive black on a roulette wheel had the same chance as a wheel that had red on its previous spin.

I'll offer up some ideas for you instead of just saying "no that wont work". If you feel so strongly about this, that your gut instinct is over riding what your brain knows to be logical, then you should keep track of paper trades or backtest your theory out. But I can save you alot of time and tell you that on all profitable backtest, sitting out days, only left money on the table, and on non-profitable backtests sitting out saved money.

If you currently use 100k BP daily and when you are up 4 days in a row and get that feeling to sit out the 5th day because you feel like your pnl is taking too huge of swings to make you uncomfortable with losing all your gains on the 5th day, the best idea I have to combat this, is to reduce your average risk to 75kBP, what this does is you will still get exposure to your positive expectancy trading yet youll be better able to with stand the deviations in dialy pnl.

 

[jedwards]

Actually, I do think you're right about this, and in fact, I'm going to do what you suggest and just trade less size with respect to my own trades. I've been suffering from too wide swings for sure.

Regardless of whether I believe you're right I'm curious to investigate this further. I know that theoretically, each trade is independent and the actual outcomes (should) have no relation with each other.

However, for example, my own particular setup on average produces 2 wins, 1 loss and 2 moved stops (small losses) per week (going back about 4+ years, but this relationship also holds true across 3 different markets). This is the historical average. Over the course of an entire year, I believe a "good" month is when I have greater than 4 more wins than losses. So, for example, if I am in the middle of month, and I've already collected 8 wins, my gut is telling me that I should stop trading for the rest of the month and book the profits because chances are that the rest of the month will just dilute the profits.

I know that we're supposed to play every trade to eliminate our own inability to know in advance which trade will win and which will lose. But it just seems as though if a particular month is having greater than expected returns early on, that just like a particular stock that has moved too much too quickly, it's bound to return-to-the-mean.

It definitely sounds exactly the same as the dubious "strategy" of betting on black when you see 5 reds in a row in roulette, since the probability is exactly the same since the next result is not affected by the previous results.

Regardless, I will see if I can backtest these "meta-trading" rules and if they are useful. I have a feeling that like you said, it will just dilute the outcomes for winning/losing strategies, but if I find otherwise, I'll let you know

 

[Engine99]

Guys, this is a great discussion. I've been toying with similar thoughts for quite some time. Since I can't easily get into/out of positions here was my approach (similar to doubling down).

I was going to put indicators (mostly oscillators like stochastic) onto my P&L (treat my P&L like a chart). if I'm in the oversold territory I was going to increase my size. Once I'm overbought I'll back off and go back to normal. Based on the limited data that I have it looked promising but overall it may result in the same behavior (lose less, win less). I can backtest this for a single strategy, but I've a tough time backtesting this for the entire portfolio.

So jed, if you can share your results, I'd appreciate it !!

 

[DGunz]

After thinking about this and realizing that not every one is trading market neutral I think that it may be possible to sit out certain days and have it make a positive impact.

The reason I say this is because; what if you have a long only system? And your system tends to do extremely well on up days, while suffering during pull backs. You may want to work in a filter of do not trade if market is up > 4 days in a row, and in a way the proxy for this is, if you are up 4 days in a row...

I do not trade directionally so I can not tell if this sitting out days would be beneficial or not. But I am just changing my mind and saying...maybe its a possibility. Youd really have to test this theory out, but I favor the thinking that if trading your strat makes you uncomfortabe with large swings the best thing to do is diversify...and if you trade futures contracts that is a very hard thing to do, since its only one security.

 

[Engine99]

Sucky trading week (down 1k1). Started out awesome with Monday (I was up over 3k) but on Tuesday I got stabbed with a knife and on Wednesday the market turned the knife in my chest. Friday the knife almost exploded when the market went straight up again but finally retraced some, so Friday's damage was manageable. Anyway, not how I envisioned this week to end.Spring break was fun with the kids, awesome skiing and both kids broke 30 skiing days this season. As far as my strategies go, S10 is a disappointment this year and I'm making some adjustments. Everything else is performing ok considering that there's no volatility and the market goes up like crazy. I can't beat the S&P 500 in this scenario but I'm actually happy I didn't blow more $$$ this week. It looked pretty bad this morning when the S&P was up 10 points. S81 (my pairs trading strategy) didn't find any trades this week which is a little concerning, it found two to three trades almost daily when I toyed around with it. S2 exited several trades this week and has only three positions open right now, S21 definitely lives up to the high expectations so far Next week work starts again and I'll be going to D.C. Mon-Friday, so systems will be back on auto-pilot. For all the ones following this, I hope you did better and actually didn't lose money. Have a good weekend and if you live in Colorado, enjoy the last weeks of spring skiing !!!
Here are the results in detail:
S1:-2677
S2: 7876
S10: -2389
S12: 1206
S11: 4187
S21:1502
S81: 2130
Weekly Performance:
-1085
S&P 500 Weekly Performance:
4769
Overall Performance since launch:
11835
S&P 500 Performance since launch:
15217
Risk Free Checking Account Performance since launch:
581

 

[Engine99]

Good trading week (up 1k6), terrible work week (very little sleep and several days where I had to work till 2am). Anyway, good news first Market lost, I made money and that even though volatility is basically non existent.
I finished listen to the “The Quants” audio book. Interesting book. Even though I barely had a chance to gamble I did ok @ the online poker tables this week. I hope I have finally turned the corner and am able to play like I play in real casinos.
I didn't see anything from D.C. this week but that's ok. Here are some updates. S2 basically is winding down positions rapidly right now. Only two legacy positions open as this earnings season starts. It did enter one new trade already which is slightly under water right now.
S21 is living up to the expectations, S10/11 did ok since we really didn't move this week at all. S81 opened two positions, which both look very promising so far, so I'm keeping my fingers crossed.
Overall trading is an afterthought right now, this year and I'm happy to be green this week with very little effort. I hope I finally gave up on home runs and focus on the steady returns, weeks like this are very relaxing since there are no significant swings.
Got back home today and will go to the mountains this weekend before I hit the road again next week.
For all the ones following, have a great weekend and I hope you did well !!!!

Here are the results in detail
S1:-2677
S2: 7911
S10: -1585
S12: 1206
S11: 4096
S21:1297
S81: 2547
Weekly Performance:
1612
S&P 500 Weekly Performance:
-573
Overall Performance since launch:
13447
S&P 500 Performance since launch:
14644
Risk Free Checking Account Performance since launch:
628

 

[Mike805]

Quote from DGunz:

... I suspect that the cold hard facts are that the system's expectancy does not change based on what has already happened. The tenth consecutive black on a roulette wheel had the same chance as a wheel that had red on its previous spin.

I disagree.

Serial correlation exists in many markets. For example, many (if not all) succesful volatility based edges rely on mean reverting volatility. Although not necessarily practical, GARCH is a great example from which to understand volatility dynamics and how academics and traders apply them. Autocorrelation is one of the fundamental dynamics driving GARCH.

If your system is a good one, then it is capturing a fundamental behavior in the market. Assuming you are capturing that fundamental behavior correctly, your system will very likely display periods of autocorrelation, which can be exploited.

From wikipedia:

ARCH models are employed commonly in modeling financial time series that exhibit time-varying volatility clustering, i.e. periods of swings followed by periods of relative calm.

I will say with absolute certaintity that all stocks display this behavior.

I am not saying you are using anything related to GARCH, but, I am telling you that your system likely works because it is (possibly unknowingly) using some fundamental facts about volatility. Heck, flipping a coin during high volatilty is a profitable edge... just some food for thought.

Mike

 

[DGunz]

Quote from Mike805:

I disagree.

Serial correlation exists in many markets. For example, many (if not all) succesful volatility based edges rely on mean reverting volatility. Although not necessarily practical, GARCH is a great example from which to understand volatility dynamics and how academics and traders apply them. Autocorrelation is one of the fundamental dynamics driving GARCH.

If your system is a good one, then it is capturing a fundamental behavior in the market. Assuming you are capturing that fundamental behavior correctly, your system will very likely display periods of autocorrelation, which can be exploited.

From wikipedia:

ARCH models are employed commonly in modeling financial time series that exhibit time-varying volatility clustering, i.e. periods of swings followed by periods of relative calm.

I will say with absolute certaintity that all stocks display this behavior.

I am not saying you are using anything related to GARCH, but, I am telling you that your system likely works because it is (possibly unknowingly) using some fundamental facts about volatility. Heck, flipping a coin during high volatilty is a profitable edge... just some food for thought.

Mike

I agree with you. I changed my mind. If you have volatility based strategies, heck even if you dont, likely if affects you (possibly unknowingly). Volatility will affect expectancy, so I take that back. I suppose what I meant to say was that, If your system has positive expectancy even in low volatility you should still trade it.

I run models that consider volatility so that I trade higher volume on high volatility days. I dont sit out days, because even the low volatility days my strat has a positive expectancy, therefore I trade everyday.

Jeds overall point was that he wanted to stop trading after being up a few days straight, he made no mention of volatility, but it might be none the less affecting his profitability; timing the days he trades "might" be beneficial as I admitted in my last post. I am operating under the assumption that everyone knows their expectancy even on low volatility days, it might not be high, but if its positive then do it. For example Janurary was rather low volume for me, and my daily expectancy was very low, and barely in the black after costs, but the point is I traded because it was still profitable.

 

[Engine99]

Very strong trading week (up 9k7). Stressful @ work but my trading systems fired on all cylinders. Very good week while the S&P 500 lost some money. I'm still on the road getting home tomorrow evening. Here's a quick recap. The low volatility helped S10, 11 and 21 (especially 21). S21 is trading with increased leverage right now, so that's why these results look very good. S2 just has a single position open and that position is not even full size since I didn't get fully filled. S81 continues to do well in this environment, but it also just has a single trade open right now. Unfortunately I've high exposure to vega right now and hope that we are not going to have a crazy open for next week. I'll be @ home all next week finishing my current work assignment. Next week is also season closing for skiing, so I'm sure we'll get up to the mountains one more time. Still excited about this trading week, every day was a three digit winner this week. For all the ones following this journal, have a great weekend !!!!

Here are the results in detail
S1:-2677
S2: 13996
S10: 1555
S12: 1206
S11: 4267
S21:6868
S81: 2751
Weekly Performance:
9760
S&P 500 Weekly Performance:
-1622
Overall Performance since launch:
23207
S&P 500 Performance since launch:
13022
Risk Free Checking Account Performance since launch:
674