This post was inspired by a Peter Lynch quote I read over at the Big Picture:
In this business if you’re good, you’re right six times out of ten. You’re never going to be right nine times out of ten. –Peter Lynch
The quote applies to our style of very short-term trading as well.
The graph below shows the S&P 500 (black) and 10 hypothetical portfolios that randomly picked the next day’s direction correctly just 55% of time (even less than Lynch’s 6 out of 10) since 1970:
[logarithmically-scaled]
All of these portfolios trounced the market, returning in the neighborhood of 12-20% a year (compared to about 6% for the market). This is only 10 examples, but we could do this all day long and come to basically the same conclusion.
Over a long enough horizon, we don’t have to be perfect. We just have to find enough quantifiable edges to be right a bit more often than we’re wrong.
This can be accomplished only two ways: increasing our winning % or increasing the size of our winners vs our losers. That’s it. But very, very, very small edges, exploited over long time horizons, at the end of the day equal very profitable portfolios. And that is what this blog is all about.
Happy Trading,
ms
P.S. as always, apologies to the real Quantifiable Edges for taking your namesake in vain.
To stay up to date with what’s happening at the MarketSci Blog, we recommend subscribing to our RSS Feed or Email Feed.
Filed under: Random Stuff | 15 Comments
There is a third way – stay in the market, but reduce your losses (a variant on the ‘increase our winners’ tactic). The post’s main contention is still valid, but this is a valid other means of gaining an edge. If (insted of doubling up on the one share you thought would beat the market) you declined to invest in the one share you thought would lag the market – and those signals are a lot clearer [hello circuit city, detroit, freddie, fannie, delta air etc.] than who’s best (Apple, Google etc.).
I agree wholheartedly with the post, but would expand the ways of gaining an edge to include just excluding the dead wood.
happy trading,
Nick
RE to nick: agree with everything you said. Lots of ways to accomplish increasing the size of winners vs losers. Another would be using appropriate stops/profit targets. Thanks for the comment, michael
avoiding loss is the nicest way to produce superior risk-adjusted returns over the long term. don’t forget about risk! investor john hussman is really good at that. he probably won’t beat the mkt in major up years but he preserves capital in down years and over a full market cycle (including bull and bear markets), that strategy most often comes out ahead of the mkt and can give you less heartburn. the caveat is you have to be willing to accept that friends and family will gloat in big up mkt yrs as your fund just plods along. but i don’t care about 1 yr bragging rights, it’s the long term that matters.
I’m guessing that you are doing a kind of Monte Carlo analysis but using the actual index data. If you could elaborate on how you did this (for us beginners) it would be much appreciated. Also, it would be very interesting to know what the results were with different Winning Percentages and to compare S&P returns versus the model in risk-adjusted terms.
Thanks,
Josh
RE to Josh: not nearly that complicated. Just using a random variable to correctly choose 55% of next day directions…can be done in Excel. In terms of risk-adjusted return, this hypothetical test is going to significantly outperform the S&P 500 because it makes one very unrealistic assumption, that being right today doesn’t impact the probability of being right tomorrow. This is of course not true in the real world and is the reason that the equity curves are so straight.
The important take away from the graph is not any of the above – this is just a proof of concept. The important take away is that we don’t have to be right as often as people sometimes believe. We shouldn’t beat ourselves (or others) up for a bad trade.
The only thing that matters is if we’re able to on average exploit a very small quantifiable edge over the market. I think we’ve demonstrated with our own strategies and with a number of strategies shared on this blog that that is definitely possible (with a bit of hardwork).
Thanks for the comment,
Michael
1.
“this hypothetical test is going to significantly outperform the S&P 500 because it makes one very unrealistic assumption, that being right today doesn’t impact the probability of being right tomorrow”
Isn’t that similar to the gambler’s fallacy, i.e. thinking that because you’ve had several hot hands your odds are now lower, when in fact the odds are the same for every “draw”. Assuming of course a reshuffled deck. Don’t get me wrong, if you have good reason to believe that the market you trade mean-reverts between periods of trending vs. non-trending and your system is trend-following then it makes sense that your odds of being right tomorrow would be affected by today’s results.
2.
Does this sound like a good way to replicate your study in Excel?
Let’s assume we pick the day’s direction correctly 60% of the time.
Column A = date
Column B = opening price
Column C = closing price
Column D = random number from 1 to 10
If current cell of Column C > 6 then absolute value of B minus C is added to equity.
Otherwise, the absolute value of B minus C is subtracted from equity.
I hope that makes sense. I’m sure there is an easier way to do this in excel, but I mainly design systems in Tradestation.
Thanks again, Michael.
oops, i meant if Column C > 4 then we profit, otherwise we lose.
RE to Josh: I think your comments re: the gambler’s fallacy are spot on. In the real world, the market goes through themes (such as the “women and children first” theme in October) and good and bad returns do tend to come in runs.
Your math on the simulation is basically correct. I used 55% correct trades, so in your example, you’d replace 4 with 4.5. We wouldn’t use the “absolute value” of the daily change, rather you’d do [yesterday's portfolio value * (today's S&P 500 close / yesterday's S&P 500 close)] or the inverse if it was a short trade [yesterday's PV * ((((today's close / yesterday's close) - 1) * -1) + 1)].
Hope that helps.
michael
Michael,
Thanks again, that does help. I’m glad I happened to comment when you were near your computer. :)
By the way, I love the blog. I’ve been reading it since you started. We need more trading systems oriented blogs. Well good ones, I mean. While I love the historical studies presented on Quantifiable Edges, Vix and More, and Traderfeed, they don’t have as much of a “system development” theme as you do. Keep up the good work. :)
Josh
Okay, I swear, after this I’m not blowing up your comments anymore tonight.
What’s up with the AdaptiveTradingSystems.com link in your sidebar? It is password protected, even on the homepage. It sounds interesting; is this a private forum or service, and is there way to get access or more info?
Thanks,
Josh
That’s strange. You mean, when you click on the link it requests a password? Hmmm…doesn’t do that for me. It’s a new site and I’ve noticed that a couple of the links are buggy. I’m going to drop it off the bookmarks list until they get some of their troubles worked out. But yes, the concept is interesting. It’s actually not as complicated as it sounds (very logical once you consider it), but still interesting.
michael
Weird. I searched for it on google and got no password request. But clicking from your sidebar does.
Alright, I lied. another comment…
Have you used any of their software?
RE to Josh: nope…I understand the concept that they are using (and have used similar ones myself), and I think it has value, but I would rather code the testing mechanism myself. Control freak I guess.
Michael
[quote]What’s up with the AdaptiveTradingSystems.com link in your sidebar? It is password protected, even on the homepage.[/quote]
Ummm… that was my fault. To cut a long story short, I accidentally created a htpasswd file in the main directory (cringes). I thought it was there for a couple of minutes or so before I realized. Maybe it was cached. I’m not aware of any links that are playing up. I did rework my blog yesterday as well, so that may have caused some problems.
[quote]It’s actually not as complicated as it sounds (very logical once you consider it), but still interesting.[/quote]
Watching the swarms drift to new locations within the parameter value / performance space is fun as well as interesting! It’s another way to increase the ability of your models to adapt. PSA is useful when model parameter values are drifting, but it is no good for binary parameters. I’m looking into incorporating something along the lines of a GA for binary inputs.
James