This week I’m going to be looking at TradingMarkets’s 10 Trading Rules (edit: read the entire series).
While I’d usually be wary of taking trading advice from any site with that many flashing banner ads, many of the rules on TM’s list are similar in spirit to concepts I’ve talked about on this blog and very much against traditional investment thinking.
Rule #1 – Buy New Lows, Not New Highs
TM looks at two alternative strategies: (a) buying when the market reaches a 10-day high and selling when it crosses below its 10-day moving average, or (b) buying when the market reaches a 10-day low and selling when it crosses above its 10-day MA.
Below, I’ve reproduced both alternatives from 1993 (the period TM’s article covers) trading the S&P 500 index:
[logarithmically-scaled]
The first strategy, buying at 10-day highs, is akin to what most individual investors do – chasing the winner until it’s not the winner anymore, and the second, buying at 10-day lows, is akin to one of the ways we approach the markets – buying where the crowd isn’t (yet).
Geek note: this is a proof of concept so these results are frictionless (no slippage or transaction costs). TM’s article assumed a trader used ETFs and intraday orders, but that’s not really what I do (I trade leveraged mutual funds), so I’ll assume an investor only bought/sold at the close. Slightly different approach, same conclusion.
And for the number-lovers:
Clearly, since the early 90’s taking a contrarian approach to the market (in this intermediate timeframe) has been far superior to a momentum-chasing one (which I think is a big reason why individual investors perform so poorly on average).
But what TM doesn’t mention is that, for a good part of the market’s history, the opposite has been true. Prior to the early 80’s, such a contrarian approach would have disastrous. Same test results from 1951 below:
[logarithmically-scaled]
I am not invalidating TM’s rule. I wouldn’t trade the rule specifically (because I think there are much better ways to time the market in this intermediate timeframe), but I agree with it 100% in spirit. It’s a core principle behind a lot of what I do.
My point is that (like daily follow-through) this very basic characteristic of the market has evolved, and ultimately, trading strategies should be designed to adapt with it.
More to follow on this week’s TM-inspired series.
[Edit: click for a summary of all related posts in this TradingMarkets series]
Happy Trading,
ms
P.S. My opinion only extends to what I call intermediate indicators. I think long-term indicators like 50/200-day moving average crossovers are a very different animal. More on this when I get to rule #3.
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: Evolving Markets, Trading Strategies | 22 Comments
-
Subscribe to RSS Feed
Subscribe via Email
My email: michael @ marketsci.com (remove spaces)
You are a blogging fiend!
Good stuff. I like TM’s concepts.
What I find amazing is how poorly buying new lows has performed in the past, and how WELL that strategy has been doing over the last decade or so. Some of the buying weakness and selling strength strategies are literally going parabolic.
What I would like to see discussion about is how long can we expect these strategies to continue to work going forward? Since they didn’t work for 30+ years, I’m wondering if it isn’t reasonable to assume that they could continue working for another decade or two? Probably not a reasonable assumption, but something to consider, nonetheless.
Funny – I’ve been working on a similar series based on the Larry Connor’s book review that I wrote. You will find this to be a theme throughout his book – namely – that he starts his tests in the 90s. But I agree with your premise that the system should adapt.
RE to Wood: you just gave me an idea for a study that I should have done a long time ago. Here’s my thought – there are a number of indicators that went through this evolutionary shift. Off the cuff I feel like shorter-term indicators (like daily FT) evolved more recently, while intermediate-term indicators (like this one) evolved further back. So I think perhaps it would be a good idea to take all of these known shifts look at when they happened relative to each other. Perhaps that will help flash the warning light when (not “if”) they begin to evolve again in the future. I like it. More to follow. michael
Agreed – an interesting idea – something as simple as a moving average of returns from each strategy and checking whether it is rising or falling.
I would test the indicators as far back as you have data. . .
Mebane, I have tested these indicators, some as far back as 90 years on the Dow, and almost 50 years on the S&P. This stuff did not work, for a long time. However, most of the patterns are the same…long period of severe underperformance, a few years of sideways returns, and then BOOM, they are like racehorses launching from the starting gate.
Michael…good idea. It would be unlikely, but possible, that they begin to fall apart in the opposite order of when they began to work. Maybe?
Damian, good idea, and it is one that I have been starting to work on. Unfortunately the maintenance required on that type of project is time-consuming. Could be extremely valuable data though.
For individual securities, it’s ironic that the positive momentum strategy has also worked over the same time period as this contrarian strategy. The two strategies are different yet I think both show promise. The positive trend following system by Blackstar funds is detailed in-depth here and their research seems fairly airtight over the sampling period of 1983-2004:
http://www.blackstarfunds.com/files/Does_trendfollowing_work_on_stocks.pdf
RE to scott: thanks for the link. Generally, I don’t research indivdiual stocks, but speaking for the broader market, I agree it trends at very long horizons like this. The 10,000 foot view (again, for the market as a whole) is that it is contrarian in short and intermediate timeframes and trend/momentum-driven at longer timeframes. Thanks, michael
Eric Crittenden has done some interesting work on momentum investing. In the article you can read it at blackstarfunds.com, he tests an ATH strategy.
The findings run counter to this rule, which I have little faith in. I have designed profitable systems using the longer term MA combinations mentioned above btw. In this case the 50/200 cross produces acceptable returns over 20 years of data and over 40 markets of stocks, cmdtys and currcies. However, it requires a certain money management scheme to be profitable.
RE to Henrick: if you’re referring to the Blackstar report on buying new all-time highs, then you’re talking about something completely and entirely different. As I tried to make clear in my “P.S”, I believe that the market trends in longer timeframes (which is why something like a 50/200 crossover strategy does produce okay returns), but that it is contrarian in intermediate timeframes.
You’re lack of faith, while appreciated from an anecdotal perspective, is just that anecdotal. The numbers are the numbers. I don’t make them up, I just report them.
michael
I coded Larry Connors “double 7s” strategy in Matlab, which is: buy at close of a 7-day low, and sell at close of 7-day high, but do this only when the price is above the 200-day MA. As he did, I backtested $NDX from Jan. 1995 to April 2008, and I checked several other ETFs. For NDX, the result was 15% annual average return, and 77% of the trades (113 trades total) were profitable. Also had good results on GLD, FXI, EWZ where the % of trades profitable was 65-75%.
Next, I modified double 7s (per Elder’s methods) by
1. set buy-stop above close of 7-day low, and keep moving buy-stop down above each day’s close until you are bought in;
2. set sell-stop below close of 7-day high, and keep moving it below the close each day until you are sold out;
3. One important exception on #2: After 7-day high (day 1), if the next 3 days’ highs finish above day 1’s close, loosen the sell-stop to just under the 7-day low band.
Results for NDX: 42% av annual return, 70% of trades were profitable. I call that pretty darn convincing it’s a winning strategy.
I can backtest any stock as far back as you wish if yahoo finance has the data.
I’m guessing that the difference in this article is he didn’t sit out when price was declining, i.e. price was < 200-day simple moving average.
Kirk Dolan’s analysis looks intriguing. Any insight on whether it works to trade any of the inverse ETFs that way? Most are well above their 200-day moving averages now.
RE to John: I talked a little bit about accounting for the broader trend in this post (another TM test):
http://marketsci.wordpress.com/2009/01/07/testing-tm-rule-3-and-4-don%e2%80%99t-fight-the-long-term-trend/
I haven’t tested Kirk’s rules specifically (because I don’t use intraday orders), so I’ll leave it up to a savvy reader to leave some thoughts on this one.
michael
I retested NDX and found a small error in my code, so the av. annual return for NDX was 34%, not 42%. 76% of trades were profitable. Sorry for the error.
We don’t have many years of data on the inverse ETFs. However, for DTO, double 7s works well. SKF is an exception. It is so volatile, that one would have to use something like a 4-day price channel. I can check others if you have one in mind.
Kirk,
I read your reply to Larry’s Double 7 in S&C mag and found you on the web. I have a question on your rule number 3.
3. One important exception on #2: After 7-day high (day 1), if the next 3 days’ highs finish above day 1’s close, loosen the sell-stop to just under the 7-day low band.
It doesn’t make sense to me when you say the next 3 day’s highs finish above day 1’s close. First I’m assuming this is your first exit , day 1. If this is the case then if you have your exit price set at the close-0.1 then what does this have to do with the highs? I’m not following this. I also don’t understand the low band statement either. Would it be possible to post a code example of this rule.
Thanks,
Cyrez
Kirk,
Like Cyrez above I need a little clarification on your rule number three. My confusion is about the lowering the sell stop below the low of the seven day band.
Am I correct in assuming that by relaxing the sell stop, sometimes you are stopped out at a smaller profit or break-even in exchange for further profit potential?
What does the modified system return without observing rule number three. Could you post results for that scenerio?
Thanks for your time,
Brian
Brian is correct. All I’m doing is seeing if the stock shows some strength for 3 days after hitting the 7-day high. If the high for those 3 days (days 2,3,4)stays above the close on day 1, then on day 5 I loosen the stop to give the stock room to run-up. Results indicate that catching many of these run-ups gives better average annual performance than automatically selling on the 7-day high. I can post other results next week.
Kirk
Very interesting Kirk. Any chance that you could send the Matlab code you are using and the results you get on other ETFs…
How does the same system on QQQQ compare with your result on the NDX… sometimes the difference can be surprising
My e-mail address is amn-inv@gmx.net
Best regards from Sweden
Dan
Kirk,
I tried coding your strategy into Fido’s wealthlab. I’m finding that it does beat buy and hold from 1995 through today, but you still get a big drawdown (around 27% in 2000. The part I’m still confused about is the 3-day rule. I’m finding that if I raise the sell-stop so that it is 10 cents below the previous days close after the 7 day high was initially reached, that I’m ALWAYS getting stopped out before a 3 day run-up is finished. I’ve tried waiting 3 days after the initial 7 day high was reached (without raising the initial stop) and then deciding whether to raise the stop or using the relaxed stop method based on whether a 3-day run-up happened. However, it doesn’t result in better returns. What am I doing wrong?
Another question: Is a 42% average return the same thing as a 42% compounded annual return?
Thanks.
Compound Annual Growth Rate (CAGR) and average yearly returns are not quite the same. For CAGR, you take the starting and ending account value over the entire period (often many years/decades) and then calculate what the return would have been to get from the starting to the ending account value if the return had been smooth/consistent throughout the entire period. In effect, CAGR doesn’t care what the actual returns were in individual years, just the growth over the entire period and how many years were in the period. If you want a formula to calculate CAGR, see here: http://www.investopedia.com/terms/c/cagr.asp
I meant to say “raise the sell stop so it is 10 cents below the next days close”. Main point is that I always get stopped out before a 3-day run-up is finished. What am I doing wrong?
Oddly enough, Yahoo data for ^NDX gives worse results than Fido’s .NDX. I don’t know which of the two is incorrect (I haven’t tried to determine that).