Use access key #2 to skip to page content.

EverydayInvestor (< 20)

New evidence against market timing

Recs

22

February 20, 2009 – Comments (13)

See original synopsis on the CXO Advisory Group's blog and see the original academic article ("Data Snooping and Market Timing Performance") that just came out.

Conclusions of the paper: 

"In isolation, over the entire 1980-2007 sample period, the best-performing simple (complex) rule outperforms a buy-and-hold strategy by an annualized 2.93% (3.18%).

Over the the full sample period, no simple or complex market timing rule significantly outperforms a buy-and-hold strategy after correcting for data snooping bias.

Over 1981-1994 and 1995-2007 subperiods, no simple market timing rule significantly outperforms a buy-and-hold strategy after correcting for data snooping bias.

Over the 1981-1994 subperiod, certain complex market timing rules do outperform a buy-and-hold strategy even after correcting for data snooping bias and accounting for reasonable transaction costs.

However, over the 1995-2007 subperiod, no complex rule significantly outperforms a buy-and-hold strategy after correcting for data snooping bias."

For those who are not statisticians, data-snooping is what happens when multiple researchers look for patterns in a large set of data. With everyone looking for ways to time the market, people are bound to find ways that would have worked in the past, just by chance alone, if they look hard enough.

Also, for those of you who say, "but I've been using X market timing system for 3 years and it works," you are falling prey to a failure to use a strong self-sampling assumption (explained in the Wikipedia article on the Anthropic Principle): "Each observer-moment should reason as if it were randomly selected from the class of all observer-moments in its reference class." Instead, you are reasoning as if the market crash had to happen and your market timing system would have worked to predict it.

Of course your market timing system has worked in the last two years! This has been one of the worst periods ever in a stock market! Any system that was not fully invested in stocks would have outperformed the market, no matter how random or useless it was. For example, if I decided to go to T-bills every time a movie with at least two "A"s in the title hit $50m in box office receipts, and got back in when a movie with two "B"s hit $50m in box office receipts,  I would have outperformed the market! Yet when the stock market rises again, all market-timing systems will underperform (because they will hold cash at various times). And can you be so sure that your system will do decent enough to keep a future bull market from wiping out the system's relative bear-market performance?

I find it amusing that everyone is buy-and-hold at the top of the market and a market timer at the bottom. Even assuming only moderate population and productivity growth in the United States, our stock market is posed to offer very nice risk-adjusted returns in the next twenty years.

 

13 Comments – Post Your Own

#1) On February 20, 2009 at 11:06 AM, EverydayInvestor (< 20) wrote:

Let me add an analogy to bolster my argument in the last three paragraphs:

Imagine you are a lonely guy (not blessed with the perfect wife, unlike me) looking for a date. You go to a bar / church / cafe / anti-bailout protest and you ask a woman out on a date. You use a funny joke / pickup line. She says yes and agrees to go on the date with you. If you conclude that the pickup line is what did it, when previous pickup lines never worked, you draw the wrong conclusion. What you need to do to conclude that the pickup line worked is to ask other women out using the pickup line and see if it works then. Otherwise, you are confusing luck (or maybe just good chemistry, shared interests, etc.) for the success of your pickup line.

Likewise with market timing in a downturn. The downturn makes everything look good that wasn't in 100% equities. You need to test your system (out of sample, of course) in a bull market to be sure that it actually works.

The Witch Doctor (hmm, I think that witch doctor made the same mistake)

Report this comment
#2) On February 20, 2009 at 11:17 AM, abitare (43.81) wrote:

Everyday,

I think u had the top Bull pitch on the India ETF, I think I had the Top Bear Pitch.

So I found a tribute song for your Pitch:

Also take a look at WFC - I am short here

Take a look at SWHC - long here

Sold my FAZ today. I am mostly cash now.

Also what software are you using to day trade?

Thanks in advance.

Report this comment
#3) On February 20, 2009 at 11:26 AM, Paxtor (29.00) wrote:

nice to see someone else long on Smith and Wesson!

Report this comment
#4) On February 20, 2009 at 12:01 PM, EverydayInvestor (< 20) wrote:

abitare - I use RediPlus through Goldman (with a secondary clearing acct at Penson) and Interactive Brokers.

I don't know enough about the banks to be short here and I don't go long any individual stocks anymore--only do short-term trading.

Report this comment
#5) On February 20, 2009 at 12:10 PM, beyondanonda (27.80) wrote:

Everyday

By extension, since the vast majority of fund managers cannot outperform the S&P, putting money in  large baskets  is perfectly acceptable.  Identifying bull conditions and being long then taking money off the table or being short during bear markets has been most rewarding to me.  Of course my gambling nature insists on some speculation,  but identifying macro market trends and holding has been my mainstay.

Report this comment
#6) On February 20, 2009 at 4:43 PM, lenri (71.53) wrote:

Market timing only works well with high beta companies or in extremely volatile periods where many boring stocks suddenly become beta plays. I never owned companies over 12 months unless they dropped like a rock in the past because the info on them was bad or the macro changed in their industry. I like it best when I can set my target on an undervalued stock, make a cozy 20-25%, and the sell while it is still a tad undervalued. Unfortunately the market today is so treacherous that for the first time, in real life, I am buying and selling on swings (not now, of course--no swing on a nosedive) making sure that the company is relatively healthy, pays a decent dividend (in case I get stuck with it) and is not on the extinct species list. I think it is a FOOL's game (no pun intended) but it is the only game out there right now.

I wish I had a million bucks right now. I'd buy every J&J-type undervalued stock and sit back for 3 years and do nothing. I know that I would double the $$. Of course what is the % of investors who were mostly in cash 3 months ago or today (assuming they had any cash left). I know there are some out there but it weren't this fool.

Report this comment
#7) On February 22, 2009 at 7:42 PM, anchak (99.86) wrote:

With all due respects to CAPsters (and Stars) ....I read the comments on the blog.....not too many people chose to critique the paper itself  - which in my view is OUTSTANDING!

Michael ....can't thank you enough for the link - however, my friend I am going to leave you with a question

Page 6:

In step 1, for each timing rule k = 1; :::; l, we generate a resample of ffk;t+1gt=Tt=R by drawing(geometrically distributed) blocks from the observed return series, with mean block length 1/q.

Have you visually considered the import of this statement ( Incidentally, Bootstrapping is one of the most robust statistical resampling methods around - however it critically depends on the method of the resample - ie for it to produce defensible replica of something that would be representative - and just so you know I speed read Politis and Romano's paper on the block resampler also - which treats the issue of stattionary resampling on a time-series )  - that it may depending on the resample put together ( on a 20 period , q=0.05) a S&P price path that could potentially juxtapose...1987 crash, followed by the highest gain, and alternating.

The problem is - By doing the resampling itself we are bringing in the notion that Market Timing should be able to handle ALL RANDOM perturbations - while Timers will argue that the series is NOT RANDOM to begin with!

Also I think one also needs to consider the variation issue - I think one of the potential value adds of Market timing - could be not Return but actually Risk Minimization ie whether the contigous sigmas were lower and statistically better.  

 

Report this comment
#8) On February 22, 2009 at 8:05 PM, anchak (99.86) wrote:

Also there's another thing which is bothering me....

Whether the resampling was done on the return series itself ( which it looks from the paper) would be the same to the Re-evaluation the rules on a Resample of the underlying series ( ie S&P % changes) - basically in math terms are ALL the rules DISTRIBUTIVE?

The reason I say this is of course say a 50day MA/200 Day MA crossover rule will trigger based on the last 200 period movements and -201th day starting value.... and say the effect ( ie rule stays on ..value 1) for exactly 20 days....the issue is if I resample the 20 day RETURN PERIOD ( ie the window in which the rule stayed on) and juxtapose that with a return period in which it didnt - However the UNDERLYING S&P SERIES created may not trigger the Rule at all. if I recomputed the 40 and 200 MA of THAT series....

 I hope this made sense!

Report this comment
#9) On February 22, 2009 at 9:55 PM, EverydayInvestor (< 20) wrote:

anchak - I get your first criticism, not your second. I just glanced over the paper. I'll take a fuller look at it.

Report this comment
#10) On February 22, 2009 at 10:51 PM, anchak (99.86) wrote:

I think I am wrong on the 2nd one - incidentally wrote the author for that clarification. Specifically Appendix B - where they mention computing the MA measures and the 80th/20th percentile over 250 day periods -

basically these measures will NOT BE stationary on a 20 period random resample - so each of them needs to be recomputed again.

 May seem a little ludicrous to you - but essentially I could simply compute all the triggers of the rules - compute the "OBSERVED RETURN SERIES" and just keep resampling the series. Very easy mathematical program - which means nothing!

The other one - involves a lot of computing time - because everything needs to be recomputed on every iteration.

I sincerely think - they would not employ the shortcut!

Report this comment
#11) On February 23, 2009 at 10:58 AM, EverydayInvestor (< 20) wrote:

ah, I get it. Yeah, I would hope they would recalculate the moving averages.

Report this comment
#12) On February 23, 2009 at 5:39 PM, OleDrippy (36.70) wrote:

#8)  "I hope this made sense!"

 Umm, no... Which is why I will be buying and holding, thank you... And just think, there are supercomputers doing what you just said in microseconds.. I'm not even going to try to come up with a "system" that can compete with that... No way.

Report this comment
#13) On February 27, 2009 at 10:59 AM, BenGriffin71 (28.23) wrote:

OleDrippy,

So that's it?  You've given in and relegated the individual as forever inferior to supercomputers when it comes to this game?

Are you really so convinced that humans are no match for computers in multidimensional games of power, greed, space and speed?  You know a super computer did recently beat a human Go master.. finally. But the super computer still required a 9 stone handicap, and was beaten the three games prior.

Can you think of some other spacial-visual pattern and trend recognition feats that barely requires a second of thought for the average human, but that event he best super computers cannot reliably perform?

“Never give in, never give in, never; never; never; never - in nothing, great or small, large or petty - never give in except to convictions of honor and good sense”

 ..... at least that was what Churchill said, but maybe if Hitler has supercomputers, he would have been a defeatist too.

 

Report this comment

Featured Broker Partners


Advertisement