Use access key #2 to skip to page content.

# bigpeach (27.64)

## bigpeach's CAPS Blog

Recs

### 36

June 19, 2009 – Comments (56)

Yesterday I was perusing the blogs and came across a comment I've heard too many times. An individual was citing another individual's CAPS score and accuracy, implying that the values in some way lent credibility to the opinions they were voicing. It's quite distressing to read things like this, not because I care what people do on CAPS, but because I worry that people may make decisions with their hard earned money based only on recommendations from Top Fools. I'm going to explore in detail the accuracy component of the CAPS ranking, but first a note on score.

Browse through the profiles of the leader board and you'll see two tickers on almost every single one. GMGMQ.PK (yes, that's GM equity), and PXCE.OB. It is near certain that GM equity is worthless, and PXCE is a shell of a company that has most likely had it's price manipulated by "pump and dumpers." So, to rack up points, you simply red thumb these (in some cases more than once) and it's an easy 100 points plus an additional "correct call." This sort of thing is clearly not actionable in practice and so is irrelevant from an investing stand point, yet in some people's mind, the points it generates lends credibility to one's arguments. More important than a person's score is how they have achieved such a score. Have they made picks that are actionable in practice, and supported their pick with thoughtful commentary, or have they gamed the system?

Accuracy

Accuracy is critically important to CAPS scores, but is it really important? I'll often hear comments like "individual X has been right on Y% of their calls, therefore they should be listened to." Oh really? I will submit that it is not at all relevant, and in fact, a computer could beat anybody's accuracy. Let's explore.

It is well known (well enough that I'm not going to source this statement) that stock prices typically follow a random walk about some longer term trend. With a margin of "success" of only 5%, how well would you do if you simply made picks at random and closed them once the score exceeded 5? To do this we simply need to run a random walk simulator and compare the results to what human players have done.

Methodology:

Lognormal daily stock prices projected for 252 days, or approximately one year (not attempting to model intra-day). The price is not in absolute terms, but in terms of price relative to the S&P 500, so it begins at 1, and a price of 1.05 gives a success on an outperform pick. The pick, either outperform or underperform is chosen at random (this part isn't actually relevant). Volatility is fixed throughout the projection. I projected 10,000 picks, and closed each as soon as it generated a score greater than 5. No pick is closed with a negative score. Esentially, I'm trying to replicate with a simulator the kind of strategy most of the top CAPS players use.

Clearly the volatility assumption is critical. The higher the volatility, the more likely it will at some point be successful. So what is our assumption? I looked at three widely followed companies, and took a standard deviation of the difference in their daily returns and that of the S&P 500. Since prices in our model are relative, so too should volatility. The period of time is 1st quarter 2009.

BAC - 11.4%

XOM - 1.3%

AAPL - 1.9%

Results

A 2% volatility during this period seemed like the most reasonable assumption to me for a real life (not CAPS) portfolio. Clearly in CAPS you can take on more risk than you would with your brokerage account. The results may surprise some people. Fully 85% of the picks were recorded as a success. That means, if you simply picked stocks at random, at some point, 85% of them would have beaten the S&P by 5%. Increase volatility to 5%, and the success ratio is 92%. If we use BAC's volatility, which in truth is the kind of pick a lot of CAPS players have been using over the past 9 months, (think triple levered ETFS like FAZ and FAS) the success ratio is a whopping 95%. So, in actuality, CAPS players, even the best, are underachieving the accuracy a computer could generate. Chalk it up to the picks they actually believe in and hold for a longer period of time.

I'll conclude by saying what I think everyone already knows, but perhaps at times forgets. When it comes to investing, it's not what you do, but why and how you do it. The Fool with a rating of 99 point something, score over 10K, 85% accuracy and average score of 5 is doing nothing more than playing the CAPS game, and should not be taken any more seriously than the Fool with a score <20. The strength of a person's research and argument is what's really important, and is the true value of this site.

#### 56 Comments – Post Your Own

#1) On June 19, 2009 at 1:57 PM, jddubya (< 20) wrote:

Good post - it is so childish when one member calls anothers score a way to rate knowledge/investing/success.

You can't judge a book by it's cover.  Or a movie by it's previews.

On a side note - "There Will Be Blood" was one of the worst movies I'd ever seen relative to the previews for it.  Spiderman 3 takes a close second.

Report this comment
#2) On June 19, 2009 at 1:57 PM, portefeuille (99.66) wrote:

great post! you are right that anyone could find this easily, so I just add the link to the relevant wikipedia entry on random walk.

Report this comment
#3) On June 19, 2009 at 2:07 PM, anchak (99.86) wrote:

Bigpeach.....You get the rec! but not straight the nod..... I have seen Hans also mentioned and echoed your argument.....

However, the formulation is what drives your conclusion - let me suggest an alternative.......

Here's my latest blog: Specifically see a very nice paper posted by Tasty over there Comment # 11.

There they are fitting a GARCH-M null model to the indices. Which means the main point there is

Volatility by itself is serialy-correlated and is a stochastic ie time varying entity.

Also given the fact that CAPS scores accuracy as an Index delta - you are dealing with 2 separate Volatility ie sigma which are time varying ( one for the index and the other from the stock).

The basic premise which you bring to the table is obvious - that CAPS accuracy should not be treated as a Coin-Toss ie 50/50 random basis - the threshold for being Non-random successful in picks is possibly different - especially if Picks are held over a long enough period.

If you can use a correct enough premise - the results of the simulation would be constructive - and it would differ period-to-period.

Report this comment
#4) On June 19, 2009 at 2:08 PM, anchak (99.86) wrote:

Hans....how come you are not researching this a little more deeply.....I am surprised

Report this comment
#5) On June 19, 2009 at 2:20 PM, RonChapmanJr (33.19) wrote:

You may have been reading one of my comments.  I have said a couple of times recently that I think GoodVibe's score is showing and will continue to support the idea that EWP is nonsense.

However, with this post I agree and disagree.  While a high ranking may be meaningless it is not the same as a <20 score.  Players who have a really low score and accuracy and have been participating on CAPS for a long time (6+ months) either are purposely trying to do a poor job or just suck at investing.  I don't care how well someone analyzes a company, if they never make money (or CAPS points) doing so they should be ignored.

ron

Report this comment
#6) On June 19, 2009 at 2:21 PM, portefeuille (99.66) wrote:

Hans....how come you are not researching this a little more deeply.....I am surprised

Because I thought someone would do it for me. And bigpeach said he was working on it. I might join in. I guess a slightly different approach that might also have a good "shot" at convincing the "critics" would be to take the picks made by say the current top 100 players and use the historic data (closing prices). The only problem would be the 200 pick limit. But I guess doing it without that limit would not be too much of a problem. So just use the calls that where actually made and use the exact starting day and price, but ignore the "ending of calls", let the computer do the ending just like bigpeach did. I hope people get my point, I could elaborate. I think foolsmethrice is also "well equipped" to do that simulation (see comment #30 here).

Report this comment
#7) On June 19, 2009 at 2:26 PM, portefeuille (99.66) wrote:

... take the picks made by say the current top 100 players ...

Not throwing them together of course. That would be our "empirical universe" so to say. A separate simulation would be done for each player.

Report this comment
#8) On June 19, 2009 at 2:29 PM, portefeuille (99.66) wrote:

I know some Monte Carlo simulation guys who could do that overnight, I guess. I am more of a pencil and paper theoretician ...

Report this comment
#9) On June 19, 2009 at 2:32 PM, portefeuille (99.66) wrote:

I am more of a pencil and paper theoretician ...

I guess Jim Simons is better off without me (see this post). I would screw up the first day.

Report this comment
#10) On June 19, 2009 at 2:35 PM, awallejr (77.67) wrote:

Here's a question I have asked several times.  Perhaps someone here will try.  If I picked a stock say at $15. This stock pays$3 per year "dividend" (alot of mlps do but while we call them dividends they are technically a return of capital).  5 years and 3 months go by. I received $15.75 in dividends. The stock is selling for say$ 10 and the S&P has gone up say 100%.  How many points would I make off the pick?

Report this comment
#11) On June 19, 2009 at 2:39 PM, bigpeach (27.64) wrote:

Yeah, I thought you might appreciate the post porte. You know, now that I finally did what I said I was going to do.

Thanks for the comments anchak. For those who may not be familiar with asset pricing; the lognormal model is a standard model for prices, and is what is used in Black-Scholes option pricing. The GARCH-M is probably more commonly used in financial modeling today. It includes error terms that are a function of the prior period error with a stochastic (random) component. It attempts to model prolonged periods of high or low volatility, as well as uncertainty. I believe it also includes price jumps (the gamma term) but I'm not positive about that.

In any case, I will respond by saying I'm trying to prove a point, rather than provide accurate numbers. A more accurate model would not provide meaningfully different results. Accuracy of 80% plus can be gamed, and should not be a factor in asessing an individual's credibility.

Report this comment
#12) On June 19, 2009 at 2:39 PM, bigpeach (27.64) wrote:

Yeah, I thought you might appreciate the post porte. You know, now that I finally did what I said I was going to do.

Thanks for the comments anchak. For those who may not be familiar with asset pricing; the lognormal model is a standard model for prices, and is what is used in Black-Scholes option pricing. The GARCH-M is probably more commonly used in financial modeling today. It includes error terms that are a function of the prior period error with a stochastic (random) component. It attempts to model prolonged periods of high or low volatility, as well as uncertainty. I believe it also includes price jumps (the gamma term) but I'm not positive about that.

In any case, I will respond by saying I'm trying to prove a point, rather than provide accurate numbers. A more accurate model would not provide meaningfully different results. Accuracy of 80% plus can be gamed, and should not be a factor in asessing an individual's credibility.

Report this comment
#13) On June 19, 2009 at 2:40 PM, checklist34 (99.73) wrote:

peachie, you are one of CAPs best bloggers, and this is another good post.  I never did get to read your thesis on GNW because I think you went on vacation before posting it.

Average gains per pick is by far something the CAPs score should include, and I don't know how i feel about double shorting something to gain more points.  you could add to a short position all the way down and make more money than the value of your initial short...  but if on CAPs if you short a 2nd time and lose you lose, the losses aren't compounded like they would be in real life if you kept adding.

And to mirror real life results, the CAPs game has to cosnider gains per pick much more than it does.  gaining 5 points average on 2000 picks is not likely to be very profitable in real life unless you have a large amount of money as transaction costs will kill you.

And lastly, the making-many-picks-is-one-of-the-real-keys-to-a-good-score thing is silly.

If we met a CAPs member who's portfolio had a green thumb for ASH from February 28th...

and thats it.

He'd have a score of 300 and probably a rank of 55 or something.  But in real life he'd be 4x his money +

Report this comment
#14) On June 19, 2009 at 2:45 PM, bigpeach (27.64) wrote:

Oh crap, another person calling me out for something I said I'd write. Well I better get on that one too. Although it's up about 3X from the time I said I'd write it so I'll be a little late with it.

Report this comment
#15) On June 19, 2009 at 2:48 PM, portefeuille (99.66) wrote:

The only problem would be the 200 pick limit.

Actually that should not be a problem because most of the "top100 players" do not end picks that have scored less than +5 score points, so the computer should have a smaller number of "active picks" for most of the time (hehe ...).

In any case, I will respond by saying I'm trying to prove a point, rather than provide accurate numbers. A more accurate model would not provide meaningfully different results. Accuracy of 80% plus can be gamed, and should not be a factor in asessing an individual's credibility.

I agree and there should be no need for more simulations and analysis to prove that point. Thank you again!

Report this comment
#16) On June 19, 2009 at 2:52 PM, anchak (99.86) wrote:

awallejr: Man! Such tricks......TMF adjusts your basis automatically - didn't you know that?

Bigpeach: Now you get the Favorite also with your comment! This is a very difficult problem my friend.

You understand that right - I think one easier approximation is to look at say +/- 30 days around Inflection points and look at contrarian accuracy.

Basically what I am saying is once the main series has a trend - with increasing volatility - and a pick in line of the trend - has a much more conditional chance of success.

Hope that made sense!

Report this comment
#17) On June 19, 2009 at 3:04 PM, Melaschasm (53.74) wrote:

The accuracy rating in CAPS is a joke.

However, I think the raw score has some value, since it requires beating the S&P in total.

For example if I green thumb and close BAC every time it gains five points, after 20 closes I will have a score of 100.  If there was not a delay between my close and new green thumb, had I simply held my green thumb until I gained 100 points, I would have closed the pick at the same time as my 20th close.  In both situations I have 'earned' 100 points, but in one situation I have wasted a bunch of time to boost my total accuracy.

*I realize my example ignores the ability to use the CAPS 20 minute delay to further game results, but I consider that a different issue.

Report this comment
#18) On June 19, 2009 at 3:11 PM, portefeuille (99.66) wrote:

*I realize my example ignores the ability to use the CAPS 20 minute delay to further game results, but I consider that a different issue.

Oh, I had almost forgotten about that one (see comment #10 here).

So we can summarise. At least 85% accuracy and at least 3% average pick score from the 20 minute delay thing. Mr. computer makes the top100 easily within a few weeks using some 2 or 3 random walk players.

Report this comment
#19) On June 19, 2009 at 3:18 PM, portefeuille (99.66) wrote:

at least 3% average pick score from the 20 minute delay thing.

Make that some 6%. He should use high volatility picks (as bigpeach has laid out) and with those a 4% gain (you have "free choice" here) for the start and another 2% (no "free choice" here) at the end of the pick should be no problem misusing the 20 minute delay backdoor.

Report this comment
#20) On June 19, 2009 at 3:23 PM, portefeuille (99.66) wrote:

(also see comment #11 here for that 20 minute delay thing.)

Report this comment
#21) On June 19, 2009 at 3:31 PM, portefeuille (99.66) wrote:

((I would actually like to see who uses that 20 minute thing. If someone finds the time it would be interesting to see who always "gets in/out" at the intraday low and out/in at the intraday high for his "outperform"/"underperform calls ... sorry for the interruptions))

Report this comment
#22) On June 19, 2009 at 3:32 PM, anchak (99.86) wrote:

I regret to inform you that you have not understood how CAPS works.....

Lets say you picked BAC at 5 and I think it almost touched 15.

So 300% gain.......

So according to your computation that would provide - 300/5 = ie 60 opportunities to pick and score the 60x5 = 300 points in CAPS!

There's a simple thing in finance called compounding.

If you just closed and repicked this stock at every 2.50 interval here's what it would look in RAW CAPS points ( without adusting for S&P)

7.5-5/5 = +50.

10-7.5/7.5=33.33

12.5-10/10 = 25

15-12.5/12.5= 20

Total points = 50+33.33+25+20= 133.33

And you wonder why UltraLong is #2 or something!

So yes this would be +3 ( as compared to 1) in accuracy - but almost 200 points less!

Never close a Green thumb in CAPS on which you have conviction!

Report this comment
#23) On June 19, 2009 at 3:41 PM, truthisntstupid (93.47) wrote:

No one with any sense would read comments from many blogs like this and ever again care the least about what any CAPS player had to say about anything - especially where real money is concerned.

Report this comment
#24) On June 19, 2009 at 3:42 PM, truthisntstupid (93.47) wrote:

No one with any sense would read comments from many blogs like this and ever again care the least about what any CAPS player had to say about anything - especially where real money is concerned.

Report this comment
#25) On June 19, 2009 at 4:16 PM, TMFBabo (100.00) wrote:

anchak: in your example, that's a 200% gain for BAC from 5 to 15.

Report this comment
#26) On June 19, 2009 at 4:18 PM, checklist34 (99.73) wrote:

did I read above that TMF adjusts our basis in the CAPs game for dividends?

thats itneresting, because in real life right now ihave a negative cost basis on ACAS after the $1.07 divi, lol. So i'd have like 50 cents cost basis on the CAPs game. I don't think its showing up that way but I haven't looked. Dividends, taxes, and transaction costs are 3 things that are a very big deal in real life but AFAI have ever understood, discounted on the CAPs game. Are dividends counted? Report this comment #27) On June 19, 2009 at 4:26 PM, anchak (99.86) wrote: bullishbabo : Oops! you are right of course...... So it is really 200 points as compared to 133 in the example. They absolutely account for divs AND Cap gains/Distbns too for ETFs etc. Report this comment #28) On June 19, 2009 at 4:48 PM, awallejr (77.67) wrote: awallejr: Man! Such tricks......TMF adjusts your basis automatically - didn't you know that? Anchak, yes I know that. So given my example how many points would I gain? I really am trying to make a point here. If my start price was$15, I accrued 15.75 in dividends, my start price would adjust to negative.  How can you possibly determine a pct gain then?  I think the game is broken, unless I am missing something here.

Report this comment
#29) On June 19, 2009 at 4:57 PM, awallejr (77.67) wrote:

Or if CAPS doesn't ever drop your start price to negative but limits it to 0, again how can you compute the points off a basically infinite pct gain comparison to the S&P.

Report this comment
#30) On June 19, 2009 at 5:03 PM, TMFCrocoStimpy (89.81) wrote:

awallejr,

Cost basis doesn't adjust by subtraction when a dividend is paid out.  The industry standard, which CAPS adheres to, is to treat a dividend as though it were re-invested into the stock on the ex-div date, at the price of the stock on that day.  The upshot is that on the ex-div date, you calculate a multiplier for your cost basis that is:

1/(1+div/share price)

This reduces your effective cost basis, but can never go to zero (or negative).

So, to answer your question, you would need to know the cost of your security at each date that the different dividends were paid out, and then multiply each of those correction factors against your original cost basis to come up with your current cost basis.

-Stimpy

Report this comment
#31) On June 19, 2009 at 5:49 PM, portefeuille (99.66) wrote:

1/(1+div/share price)

The formula should be

a -> a * (1 - d/b), where a is the old cost basis, d is the dividend and b is the price before the dividend is paid, so the "factor" is (1 - d/b).

Report this comment
#32) On June 19, 2009 at 5:54 PM, portefeuille (99.66) wrote:

With this x = (1 - d/b) you have

(b - a)/a = (b - d - xa)/xa.

Another way to arrive at this:

Before the dividend you have n shares with a cost basis of a/share.

You get nd/(b - d) new shares ("dividends reinvested").

So the new cost basis is na/(n + nd/(b - d)) = a * (1 - d/b).

Report this comment
#33) On June 19, 2009 at 6:05 PM, portefeuille (99.66) wrote:

#30-32 For small dividends (small d/b) the "error" is small because for small c (= d/b) you have 1/(1 + c) = ca. (1 - c).

Report this comment
#34) On June 19, 2009 at 6:06 PM, topsecret09 (44.31) wrote:

Point  well made,and well taken...... I agree with you 100 %

Report this comment
#35) On June 19, 2009 at 6:49 PM, TMFCrocoStimpy (89.81) wrote:

Porte,

Industry often uses the approx. of (1-div/price) instead of the full expression, but that can be problematic.  As you've noted, this approx. is only valid for small values of (div/pric).  Since one-time payouts (which can lead to larger values of div/price) are handled in the same way as regular dividends, the full expression for basis reduction is used to avoid errors from the higher order terms.  Though this does not happen often, handling the dividend properly from the get go avoids having to make case-by-case adjustments for large one-time payouts.

-Stimpy

Report this comment
#36) On June 19, 2009 at 6:52 PM, automaticaev (< 20) wrote:

when i chose tickers i never delete them or sell them i just leave it so...

Report this comment
#37) On June 19, 2009 at 7:01 PM, portefeuille (99.66) wrote:

Industry often uses the approx. of (1-div/price) instead of the full expression, but that can be problematic.

Actually I do not think that 1/(1 + d/b) is "easier" to use than (1 - d/b). Why would anyone use that "approximation"? Is the "correct formula" used in the "caps" game?

Report this comment
#38) On June 19, 2009 at 7:14 PM, TMFCrocoStimpy (89.81) wrote:

Porte,

The "correct formula" is used in CAPS.  The "easier" part is in reference to the implementation and operations standpoint.  Because we use the exact formulation, we can run any size dividend through the same calculation.  If we did not do it that way, then you would have to run an inequality test against (div/share) for each dividend to determine if you should use an approximation or switch and use the full formulation.  Cleaner (i.e., easier) to code and maintain one routine rather than two as a general rule, plus the added inequality and branching logic require far more cycles to process than the extra DIV operation in the full formula.

As to why the (1-div/share) approx. gets used in the industry at all, I don't have a definitive answer for that.  I like to think that it is legacy from the days before spreadsheets, and simply got engrained in the way things are done, but that is just a guess.

-Stimpy

Report this comment
#39) On June 19, 2009 at 7:26 PM, portefeuille (99.66) wrote:

#38 That was confusing. The factor of (1 - d/b) is not an approximation (see comment #32 above). The wrong formula is the one you mentioned in comment #30 above. Which one is used in the "caps" game? And the calculation of the correct factor involves just 1 division and 1 subtraction, so it should be "easy to implement".

Report this comment
#40) On June 19, 2009 at 7:43 PM, TMFCrocoStimpy (89.81) wrote:

Porte,

Sorry, thought you had corrected yourself in #33.  The formulation in #32 is incorrect, and (1-div/price) is the approximation to the correct formula, not vise-versa.  Before you immediately jump to say that #32 is formulated correctly, perform the simple sniff test on your answer:  If a dividend equal to the share price is paid out, then your formula would indicate that the cost basis is now 0, so the return on the original investment would be infinite.  Suggest you take a look at the problem again.

Cheers,

-Stimpy

Report this comment
#41) On June 19, 2009 at 8:11 PM, portefeuille (99.66) wrote:

If a dividend equal to the share price is paid out, then your formula would indicate that the cost basis is now 0, so the return on the original investment would be infinite.

The factor given in comment #32 above is the correct one. After a dividend of d = b the price would drop to zero and you could buy nd/(b - d) = \infty new shares, giving you a cost basis of zero, just the way it should be. That is the correct limit. The "return on the original investment would" not be "infinite", because the price would have gone from 0 to 0. So the return is 0.

Please have a look at comment #32 above.

With this x = (1 - d/b) you have

(b - a)/a = (b - d - xa)/xa and that is just the way it is supposed to be. The lhs is the "return" before the dividend payment. The rhs is the "return" after the dividend payment.

Another way to arrive at this:

Before the dividend you have n shares with a cost basis of a/share.

You get nd/(b - d) new shares ("dividends reinvested").

So the new cost basis is na/(n + nd/(b - d)) = a * (1 - d/b).

This should really not be that hard to see ...

Report this comment
#42) On June 19, 2009 at 8:13 PM, ARJTurgot (69.19) wrote:

I'd take it a step further and suggest that you not take anyone's posts here seriously.  The only thing I see the posts on Fool proving is that an infinite number of monkeys on an infinite number of keyboards doesn't, in fact, result in the creation of Hamlet.

Report this comment
#43) On June 19, 2009 at 8:17 PM, portefeuille (99.66) wrote:

To the readers. I apologise, this has nothing to do with this great post. It all started with comment #10 above. But by now I really have the feeling that they might not treat dividend payments the way they should here in the "caps" game.

Report this comment
#44) On June 19, 2009 at 8:27 PM, portefeuille (99.66) wrote:

maybe a simple example:

stock goes from 1 to 5 (+400%), pays dividend of 2.

1 - 2/5 = 0.6

new starting price is 1 * 0.6 = 0.6

0.6 to 3 is again (+400%).

Maybe I have a wrong understanding of cost basis. I was talking about the adjusted starting price.

Report this comment
#45) On June 19, 2009 at 8:59 PM, portefeuille (99.66) wrote:

I have found the error.

You talk about the "post div price" b - d.

1/ (1 + d/(b - d)) = 1 - d/b

Report this comment
#46) On June 19, 2009 at 9:00 PM, portefeuille (99.66) wrote:

so no problem, you should have specified whatshare price is meant in

1/(1+div/share price)

though. so "problem" solved, we are both right. have a nice week end.

Report this comment
#47) On June 19, 2009 at 9:07 PM, portefeuille (99.66) wrote:

To see that they use the post dividend share price have a look at this:

--------------------------

Okay, exactly how are we adjusting?
The exact formula is

New Start Price = Old Start Price * Reduction Factor

Where Reduction Factor = 1 / (1 + Dividend / Post-div Price)

--------------------------

(from here)

Report this comment
#48) On June 19, 2009 at 9:16 PM, TMFCrocoStimpy (89.81) wrote:

Porte,

I see where this is getting fouled up - we're using different values of "ticker price" in the formulation.  For your formulation, (1-d/b), you are using b==price prior to ex-div adjustment.  Re-arrange so that "price" is the post ex-div adjustment value and you come around to the original formulation.  Terminology is key.  Post ex-div adjustment is what was meant by "...at the price of the stock on that day" in #30, since the price opens with the ex-div date adjustment.

So why the heck would you do it that way you ask?  In practice, it has been considered easier to use the post ex-div price because it corresponds to the opening price on the ex-div date, and that is the data that comes in through the datafeeds and is also the price at which the new shares are purchased.

However, as has been said, this is a bit of a deviation from the main post, which I'd like to comment on a bit.   As TMFJake has mentioned in a few places, we've been examining the accuracy variable considerably ever since the launch of CAPS.  The random walk aspect of the measure has grown considerably over the past year+ with the increased volatility of the market, given the fixed threshold that is currently used.  A variety of strategies including variance adjusted thresholds (rather than the fixed +5%) and time weighted/time decay functions of accuracy are being re-examined, though whether they will progress beyond being retested I can't say.

BigPeach, I would suggest that you expand your simulations to include the 200 pick limitation constraint, and then generate a population of portfolios that evolve over a 1 or 2 year period.  Anytime you have a pick >+5 close it, and then restart a new pick in its place.  Track both the cumulative score and accuracy of the portfolios as they evolve over time.  If you have the tools, examine the distribution of accuracy v. score. v population at a variety of time slices.  I think you'll find that the pick number constraints have some fairly interesting effects on the mid to long term composition of the portfolio accuracy and score.  Another added feature of the simulation is that you want to work underperform picks into the scenario, which can have losses in excess of 100%, unless you want to ignore accuracy harvesting from underperforms (which seems to be one of the primary points people bring up when examining the whole accuracy question).

The only purpose in suggesting this excercise is that you seem to be quite interested in the problem, and I'd like you to get a feel for how rich the landscape is from just the addition of a few more components to the simulation (pick number and underperforms) in terms of CAPS portfolios that would be generated at random.  I concur that with the current market volatility, there is to much opportunity to use volatility to harvest accuracy than I would like the system to have.  However, and I stress this each time the accuracy discussion comes up, the number of players who fail miserably in trying to use this strategy is substantial, and few people tend to see that.  I would be quite interested in seeing what proportion of a random population you come up with that can maintain a high accuracy and high score using the pick limits and underperforms.

-Stimpy

Report this comment
#49) On June 19, 2009 at 9:18 PM, TMFCrocoStimpy (89.81) wrote:

Porte,

Dollar short and a few minutes late on my last post.  Glad we both caught the discrepancy.  Have a good weekend yourself - hope to chat more.

-Stimpy

PS - BigPeach, apologies if this got to far off-topic

Report this comment
#50) On June 19, 2009 at 9:27 PM, TMFUltraLong (99.95) wrote:

Nice.... so behold my 98.36% accuracy, take that Mr. Computer!!! lol

UltraLong

Report this comment
#51) On June 19, 2009 at 9:43 PM, TimothyVR (< 20) wrote:

*shrug*

I don't care at all about the battle of the scores. I'm too much of a newbie to grasp the convoluted details. There's a certain amount of "in house" bickering and competition that goes with the territory. But I have found a lot more than that.

The CAPS site is not just a source for ratings. Every stock has plenty of specific information and commentary that are helpful and clear. And yes, that DATA does help me make decisions about what to do with my had earned money. I prefer to find out as much as I can about a stock before I even think of buying and the CAPS discussions are very informative.

Report this comment
#52) On June 19, 2009 at 10:44 PM, Tastylunch (29.40) wrote:

fantastic post BigPeach! well except for this part

This sort of thing is clearly not actionable in practice and so is irrelevant from an investing stand point,

A lot of .Pk and .ob stocks ar shortable if you know where to look. It's all about having the right brokers ;) It  is significantly harder to do now that it was 15-18 months ago unfortunately.

Still the post was very awesome indeed! I think your criticisms are fair regarding CAPS. I think ti woud eb an issue if the objective of CAPS was ot find he betsinvestors but since it's not I guess it really doesn't matter other than to the poeple who dogpile Allstars with real money

The Fool with a rating of 99 point something, score over 10K, 85% accuracy and average score of 5 is doing nothing more than playing the CAPS game, and should not be taken any more seriously than the Fool with a score <20. The strength of a person's research and argument is what's really important, and is the true value of this site.

That may be one of the wisest things I've read in CAPS and I completely agree.

Report this comment
#53) On June 20, 2009 at 12:21 AM, rexlove (99.57) wrote:

Right on bigpeach. I'm tired of reading blogs from these CAPS players with high scores thinking they're the best thing to come along since Warren Buffet. Let's see some of these players make some REAL MONEY!

Report this comment
#54) On June 20, 2009 at 1:26 AM, Chromantix (97.67) wrote:

I have a higher CAPS rating than you, so I'm smarter, better looking, and more accurate!

(just kidding, +1)

Report this comment
#55) On June 20, 2009 at 7:40 AM, TMFBabo (100.00) wrote:

However, with this post I agree and disagree.  While a high ranking may be meaningless it is not the same as a <20 score.  Players who have a really low score and accuracy and have been participating on CAPS for a long time (6+ months) either are purposely trying to do a poor job or just suck at investing.  I don't care how well someone analyzes a company, if they never make money (or CAPS points) doing so they should be ignored.

I want to echo this comment from Ron.  I commented somewhere that it is quite possible for players with very good CAPS ratings to be bad investors in real life; that's if all 200 of their picks are for "gaming the system."  I must also add that it is pretty much impossible for players with bad CAPS ratings to be very good investors in real life.  While a player with a good rating can still possibly be a good investor in real life, a bad rating precludes that possibility.

As long as players do not game the system, I believe that score is quite useful.  It should somewhat correlate to meaningful investing skill.

A lot of .Pk and .ob stocks ar shortable if you know where to look. It's all about having the right brokers ;) It  is significantly harder to do now that it was 15-18 months ago unfortunately.

I agree with this from Tasty.  I believe EverydayInvestor was able to find .OB and .PK stocks to short in real life, and he made some good money doing so.  Maybe Tasty or someone can confirm this? I just thought I remembered reading that on one of Everyday's old blogs.

Report this comment
#56) On September 30, 2009 at 11:10 AM, OklaBoston (68.01) wrote:

As someone who had an All-Star rating around here back in November, and then watched said rating fall through the floor as a result of a lousy stretch starting in January, I'm grateful for your implication that 20< ratings such as my current one shouldn't be ignored.

Let's hope my current emphasis on earnings surprises, combined with paying less attention to stockcharts (while not ignoring them totally) keeps helping get my rating back up where it used to be. It has been doing so for about a  month now, for what that's worth.

Report this comment

### Blog Archive

##### 2009
September (2) July (1) June (2) May (2)