finding an edge...simple?
- Thread starter FRuiTY PeBBLe
- Start date
acrary,
This may be a dumb question, but I always find myself scratching my head.
Sometimes you will refer to a system that has an edge. Other times you will refer to a system that has made money every year for x years, but has no edge. I think with this second type of system, you usually say that they trade the character of the market.
Could you tell me what the difference is? If a system has been profitable for say 10 years, but trades the character of the market, what is the difference between that and a system with a real edge? If you have a system that has made money, how do you know if you have an edge or not? As in your above quote, don't you have an edge if you make money?
Thanks,
FRuiTY P.
I don't trade stocks. This is the SP500 futures.
What I was saying is I found periods of non-random behavior when measured against random entries and found that they persisted over many years (in this case all years that the SP future has traded since 1983).
Expectation = (%wins * size of win) - (% loss * size of loss)
There are two components to expectation:
1). % wins versus % losers
2). Size of win versus size of loss
Since 1983 there have been slightly more than 51%+ up days and 48%+ down days. So in the sample of 5100+ days the up versus down frequency has been close to random i.e. 50%.
However we all know the market has moved greatly in that time. Why? Because the size of the upmoves verus the size of the down moves has been much greater. You can look for nonrandom frequency of the event (up or down) or you can look for nonrandom size of moves (size of win versus size of loss).
Most people concentrate on the frequency which I have found is random. However if you look at when a large move happens verus a small move, you can limit your trading to the periods when the best opportunities are likely. For instance, if you only wanted to trade 30 min. of a day you could check all the 30 min. periods and only trade during the 30 min. period with the largest range. If you guessed wrong on the direction you could quickly flip the trade and ride it for the rest of the period. (That is if the 30 min. largest range happened consistently at the same time each day). It would also allow you to know that you shouldn't try to catch the big move during the lunchtime period.
It sounds like your system does have a directional bias though, since you exit only on close. Therefore it depends on having a result distribution that is skewed to one side (in terms of magnitude, not frequency).
It seems that if all you were predicting is a distribution with fat tails, but not any directional bias, you would trade it by entering either long or short randomly and then exiting on a tight stoploss or after a fixed time period has elapsed.
-bbc
Edge = better than random
Character of market = curve fit
If something makes money it could be curve fit or it could be a non-random event. If it's cuve fit, the return will depend on the market exhibiting the same behavior year after year. If it acts the same, you make money, if not you lose. In a real edge, the market behavior is linked to something and exhibits a strong corellation between something and the move.
I measure my edge by taking the trades and checking them versus random other possible entry periods given the same holding period. For example, if I have 50 long trades in a year with a one day holding period. I test 5000 random combinations of 50 one day buys with a one period hold and rank the random results. If my total profit from the 50 one day longs beats at least 70% of the random entries, I'm encouraged. Same test is applied to the short side trades. I do this for several years backtesting to make sure the edge is stable. If it is, then I trade it. When it starts to drop below 60%, I stop trading it and move on to another edge.
Since there isn't much chance anybody can influence the market on a weekly basis (and since I'm now retired), I'll describe my weekly edge I found for the SP market. What was it? It was funds flows were strongly correllated with near term market performance. I got the data from trimtabs.com. After massaging it, I found a predictable weekly pattern to the overall direction of the market (up or down).
Here's the weekly results using just the edge (enter on open on monday and exit on close on friday). Obviously I can intraday trade around this to improve the results, but here's the summary of taking one trade per-week since 2001 blindly (all the data I have for funds flows). From this it should be pretty obvious this is not random results. I ran the edge test I have against this for the 2+ year period and the long trades were better than 88% of all random combinations. The short side trades were better than over 90% of all the random combinations. If it ever drops below 60% it'll go in the waste basket, but right now it's a pretty easy method of making money.
Obviously the war is having a negative influence on the fund flows right now, so the results should improve once the trading envioronment improves. Even so, it's profitable so far this year.
For those who think that it takes sophisticated indicators and convoluted math to create a system, here are the results from the past 12 months on a very simple system based ONLY a daily MA for entry and exit, and it is not optimized or curve fitted. It trades on a list of 250 stocks that are scanned only for EPS and EPS growth. For you Trade Station junkies, this summary is from Investor RT. So you will need to suffer through it. And for those who think this market can't be traded, the results for the past 90 days on this system have yielded over 105 points, with 42 points in the past 4 weeks, with similar results back to 1997. I submit this as encouragement to those who think that money can't be made with a simple stock system based on a logical approach to the market. (Summary is based on 1000 shares per trade.)
Avg Profit/Trade $1,056.92
Avg Position Size $33,949.19
Avg % Gain/Trade 3.113%
Avg % Gain/Bar $1.471%
Gross profit $933,644.73
Maximum Drawdown $6,322.50
Commissions $13,065.00
Average Drawdown $1,353.54
Net Profit $920,579.73
Ratio Avg Win/Avg Loss 0.92
Total # Of Trades 871
Percent Profitable 78%
Winning Trades 688
Losing Trades 183
Largest Winning Trade $7,078.50
Largest Losing Trade $8,887.50
Average Winning Trade $1,882.64
Average Losing Trade $2,047.43
Largest Winning Long $7,078.50
Largest Losing Long $8,887.50
Number Periods Long 1843
Percent Periods Long 2.36%
Max Consec. Winners 22
Max Consec. Losers 3
Avg # Bars In Winners 1
Avg # Bars In Losers 3
Avg Profit/Trade $1,056.92
Avg Position Size $33,949.19
Avg % Gain/Trade 3.113%
Avg % Gain/Bar $1.471%
Gross profit $933,644.73
Maximum Drawdown $6,322.50
Commissions $13,065.00
Average Drawdown $1,353.54
Net Profit $920,579.73
Ratio Avg Win/Avg Loss 0.92
Total # Of Trades 871
Percent Profitable 78%
Winning Trades 688
Losing Trades 183
Largest Winning Trade $7,078.50
Largest Losing Trade $8,887.50
Average Winning Trade $1,882.64
Average Losing Trade $2,047.43
Largest Winning Long $7,078.50
Largest Losing Long $8,887.50
Number Periods Long 1843
Percent Periods Long 2.36%
Max Consec. Winners 22
Max Consec. Losers 3
Avg # Bars In Winners 1
Avg # Bars In Losers 3
Is this strategy a variation of "buy when dips by X% below the MA", applied to stocks with low EPS and high EPS growth?
-bbc
Mine too.
lindq, when you scanned for EPS stocks, did you actually do the scan 12 months ago? If you did the scan recently then these results aren't valid. I could do a scan today for the best EPS stocks and build any number of systems that would do well on those stocks in the past. If you have some way of knowing for sure what stocks will have good EPS 12 months from now, I would love to hear it.
Yes, that's pretty much what the prod5 system does. In this case I found the result distribution at times is predictable (though not the frequency of the distribution).
Yes, you could do that or flip the trade because what you're predicting is the day will have a larger range than normal.
That particular system looks for points which if touched the next day are likely to trigger a larger than average move in a particular direction with a small chance of loss. Any decent trader should be able to improve on the results by doing active trading against it.
I use the exit on close for testing because I found it too tough to figure the intraday entry, apply a stop and then create a intraday database to test the entries and exits versus random. It's easy enough to say get in and ride it all day and exit on close. Then all I have to do is measure the one day move of the system verus random one day moves in a database. I've thought about working on something more elaborate, but I'm not the best programmer and it just seems like alot of work. So far, there still seems to be lots of low hanging fruit for trading so I don't have to deal with this headache.There's probably some huge intraday edges, but I'll leave them for the next guy that wants to do the work.
BTW, I don't trade the systems... I trade against the systems. This gives me something to look forward to each day (like playing a game against the computer). For instance, prod5 only had one trade this week a buy at 861.50 on 3/27. Since it opened below it, I should have jumped in with a market order right away. By waiting and trading against it, I was able to buy at 858.25 when a nice volume divergence showed up. The system just gave me the added confidence that the market had a decent opportunity to the upside. I apply stops and use some commonsense to avoid days with these systems to avoid the large negative outliers.
All I really wanted to say in this thread was when you have one of these systems as a guide, it's a whole lot easier to trade.