Here's a statement of the obvious: The opinions expressed here are those of the participants, not those of the Mutual Fund Observer. We cannot vouch for the accuracy or appropriateness of any of it, though we do encourage civility and good humor.
Support MFO
Donate through PayPal
Fooled by Data-Mining: The Real-Life Performance of Market Timing with Moving Averages
Somewhat surprising results... If you are using 200 day moving averages or 10 month moving average (using monthly data) take a look at this paper. If you are short of time, skip and read the conclusions first...
I have recently tried an unconventional charting approach to help me with my trading decisions. Instead of charting a fund I own against its 200 dma, I chart my fund's performance against it's particular index. For example charting a Tech focus fund (PRGTX) against the Nasdaq 100 index (USNQX).
Here's what these two funds look like over the last 5 year time frame.
Two things seems to emerge for me. On a long term basis (5 year chart) PRGTX has consistently outperformed its Index and periodically, on short term basis, it has underperformed (last 3 month of this chart). This allows me to see if the manager is adding any alpha relative to the index. I am willing to stick with a manager short term, through a rough patch (right now), but don't see the value of owning a consistent under performing fund. Underperformance would push me towards replacing this fund with another better candidate or just own the index fund in that catagory.
Another unconventional charting technique I use is when taking profits. I chart my funds against my "steady eddy funds" such as PONDX. If any of my funds out perform PONDX ( I use 10% out performance as a trigger), I sell that 10% profit and buy additional shares of PONDX. One could also take profits and reallocate them to underperforming funds that they're holding long term (i.e. Precious Metals), they could further diversify their portfolio (start a new position in a new strategy), or move profits to cash for income needs. Here's PRGTX charted against PONDX for the purpose of monitoring profits.
I always had trouble maximizing my decision making using the 200 dma CUTLINE as a buy, hold or sell strategy. Ulli the ETF Bully dedicates a site to this approach which for some investor may be a good resource. Here's the site. Click on the internal site's links to see what ETFs or mutual funds are trending (make his cutline) right now.
" As with volume analysis, I never use moving averages in my trading. I find them to be much ado about nothing. You might want to read Larry Williams's The Definitive Guide to Futures Trading and his moving average studies. After extensive research with moving averages, moving average crossovers, and channels, Larry's conclusion is that none of them consistently make money. While Larry agrees that moving averages may be good trading tools, he says it is next to impossible to build a winning system solely around moving averages."
Thanks Investor. My initial reaction is, of course, defensive, since I have taken a liking to dynamic allocation and am a fan of Mr. Faber's research. And, because the article, to paraphrase Ned Davis, seemed "pre-wired":
The research presented in this paper was motivated by the following two observations. First, it is rather incredible that a no-brainer strategy can easily beat the market. Second, using the same stock price data over the same period as in Faber (2007), we find that the 10-month moving average rule is the best trading rule in a back test. Thus, it is natural to suspect that the reported performance of the 10-month moving average rule contains the data-mining bias. The data-mining bias in this case is defined as the difference between the observed performance of the best trading rule in-sample and its expected performance out-of-sample (Aronson 2006, Chapter 6). Out-of-sample performance deterioration is a very well-known problem in technical analysis. Consequently, one should expect that the future expected performance of the 10-month moving average rule will be worse than the reported performance.
Let me (aka no-brainer) read up a bit on "out of sample" analysis, before posting more. (I did forward the link to Mr. Faber's email, perhaps he too will weigh in.)
Reply to @bee: Hi Bee. I see you're mountain climbing again! Nice chart comparing PRGTX against PONDX to illustrate your trading technique. I started looking into your previous post as well on VTI and EDV. Will post better response soon I hope, since I found the divergences interesting. Thanks too for links to The etf Bully and Bob's Mutual Fund Page. Hope all is well.
Reply to @bee: What you are doing in the second chart is typical rebalancing with pre-established bands. Although you seem to be rebalancing from equities to bonds only.
Reply to @Investor: Ageed...probably not a bad time to rebalance some PONDX back into PRGTX. Right now I am finding myself taking more profits from equities and moving those into bonds. One could use a conservative allocation fund (I use VWINX) to rebalance into for a more balance (equity/bond) approach.
Basically I think Professor Zakamouline's paper says the back-tested sample is not long enough to be representative; therefore, he creates a model to simulate market and investor behavior then uses results from that model instead.
If he’s right, it implies that all back testing using actual market data is data mining, since no sample is all encompassing or infinite. And his model concludes: “…over a sufficiently long run there are no chances that the market timing strategy allows investors both to reduce risk and enhance returns.”
Something like that...
He uses Mebane Faber's 2007 paper, probably as a hook, but I suspect he could apply the same thesis to anybody's method...Professor Greenblatt's "Magic Formula," Professor Shiller's CAPE model, and Professor Swensen "Yale Endowment" approach. In fact, any method derived from empirical market evidence would be suspect. And something tells me he would come up with same conclusion...they will fail over time, as their methods are victims of luck and data mining bias. Similarly, the author would likely dispute Jegadeesh and Titman's 1993 paper, which seems to be starting point for momentum methods. Ditto for long-short strategies, risk parity, etc, etc.
Translation: stop trying to beat the market and stick to passive buy-and-hold.
Yet another volley in the never ending debate.
I'll look-up Aronson's paper, the reference that seems to have motivated him.
1. I did read through the reference that seems to have motivated Professor Zakamulin, entitled “Data-Mining Bias: The Fool’s Gold of Objective TA (Technical Analysis), Chapter 6,” in text “Evidence-Based Technical Analysis,” by David Aronson. An outright assault on what Mr. Aronson calls “subjective TA,” which is any TA that cannot be reduced to a computer algorithm and back tested. Subjective TA is “not a legitimate body of knowledge but a collection of folklore resting on a flimsy foundation of anecdote and intuition.” He puts after-the-fact TA rules in the luck category, along with “monkeys as authors, bible coders, lottery players.” Arguing that future expectations in the “out of sample” period are overblown based on linear extrapolations of the “in-sample” period, which has too few observations to be a useful predictor.
2. I do not think “Social Science Research Network” provides any peer or editorial review, despite listing names like Professor Fama and Professor Sharpe on its “Board of Directors.” The papers are supposed to have an “academic” style, but otherwise, it’s not much different than posts on say Seeking Alpha.
Professor Zakamulin has updated to include momentum methods. He's also provided me with optimal SMA for initial in-sample period. Will be posting more as I review further.
Comments
I have recently tried an unconventional charting approach to help me with my trading decisions. Instead of charting a fund I own against its 200 dma, I chart my fund's performance against it's particular index. For example charting a Tech focus fund (PRGTX) against the Nasdaq 100 index (USNQX).
Here's what these two funds look like over the last 5 year time frame.
Two things seems to emerge for me. On a long term basis (5 year chart) PRGTX has consistently outperformed its Index and periodically, on short term basis, it has underperformed (last 3 month of this chart). This allows me to see if the manager is adding any alpha relative to the index. I am willing to stick with a manager short term, through a rough patch (right now), but don't see the value of owning a consistent under performing fund. Underperformance would push me towards replacing this fund with another better candidate or just own the index fund in that catagory.
Another unconventional charting technique I use is when taking profits. I chart my funds against my "steady eddy funds" such as PONDX. If any of my funds out perform PONDX ( I use 10% out performance as a trigger), I sell that 10% profit and buy additional shares of PONDX. One could also take profits and reallocate them to underperforming funds that they're holding long term (i.e. Precious Metals), they could further diversify their portfolio (start a new position in a new strategy), or move profits to cash for income needs. Here's PRGTX charted against PONDX for the purpose of monitoring profits.
I always had trouble maximizing my decision making using the 200 dma CUTLINE as a buy, hold or sell strategy. Ulli the ETF Bully dedicates a site to this approach which for some investor may be a good resource. Here's the site. Click on the internal site's links to see what ETFs or mutual funds are trending (make his cutline) right now.
The etf Bully
Also, I'll mention Bob's Stock Mutual Fund Page which I believe also uses the 200 dma as a way of ranking mutual fund that are trending.
Bob's Stock Mutual Fund Page
" As with volume analysis, I never use moving averages in my trading. I find them to be much ado about nothing. You might want to read Larry Williams's The Definitive Guide to Futures Trading and his moving average studies. After extensive research with moving averages, moving average crossovers, and channels, Larry's conclusion is that none of them consistently make money. While Larry agrees that moving averages may be good trading tools, he says it is next to impossible to build a winning system solely around moving averages."
Hi Charles,
Like you, I am always trying to learn and improve...thanks for all your great threads.
Ageed...probably not a bad time to rebalance some PONDX back into PRGTX. Right now I am finding myself taking more profits from equities and moving those into bonds. One could use a conservative allocation fund (I use VWINX) to rebalance into for a more balance (equity/bond) approach.
If he’s right, it implies that all back testing using actual market data is data mining, since no sample is all encompassing or infinite. And his model concludes: “…over a sufficiently long run there are no chances that the market timing strategy allows investors both to reduce risk and enhance returns.”
Something like that...
He uses Mebane Faber's 2007 paper, probably as a hook, but I suspect he could apply the same thesis to anybody's method...Professor Greenblatt's "Magic Formula," Professor Shiller's CAPE model, and Professor Swensen "Yale Endowment" approach. In fact, any method derived from empirical market evidence would be suspect. And something tells me he would come up with same conclusion...they will fail over time, as their methods are victims of luck and data mining bias. Similarly, the author would likely dispute Jegadeesh and Titman's 1993 paper, which seems to be starting point for momentum methods. Ditto for long-short strategies, risk parity, etc, etc.
Translation: stop trying to beat the market and stick to passive buy-and-hold.
Yet another volley in the never ending debate.
I'll look-up Aronson's paper, the reference that seems to have motivated him.
1. I did read through the reference that seems to have motivated Professor Zakamulin, entitled “Data-Mining Bias: The Fool’s Gold of Objective TA (Technical Analysis), Chapter 6,” in text “Evidence-Based Technical Analysis,” by David Aronson. An outright assault on what Mr. Aronson calls “subjective TA,” which is any TA that cannot be reduced to a computer algorithm and back tested. Subjective TA is “not a legitimate body of knowledge but a collection of folklore resting on a flimsy foundation of anecdote and intuition.” He puts after-the-fact TA rules in the luck category, along with “monkeys as authors, bible coders, lottery players.” Arguing that future expectations in the “out of sample” period are overblown based on linear extrapolations of the “in-sample” period, which has too few observations to be a useful predictor.
2. I do not think “Social Science Research Network” provides any peer or editorial review, despite listing names like Professor Fama and Professor Sharpe on its “Board of Directors.” The papers are supposed to have an “academic” style, but otherwise, it’s not much different than posts on say Seeking Alpha.
Link to May 1 Revision:
Fooled by Data-Mining: The Real-Life Performance of Market Timing with Moving Average and Time-Series Momentum Rules