Trading System by Andrew Gibbs - Member of Larry Williams Hall of Fame

[zipzap]

RRRrrrrRRRRrrrrRRRRrrr, the self appointed forum police is back on the case. Sinner, I would suggest you have an in-depth look at Portfolio123 so you can provide in informed opinion, it' a very powerful program.

And, in case you didn't read earlier posts, when I get a spare few hours I'll be posting more detail on P123 and what it can do.

 

[zipzap]

In the next group of posts I am going to demonstrate how fundamental data can be used for isolating a group of stocks to trade based upon a combination of fundamental and technical considerations.

The first step is to build a ranking system

For this post we'll have a look at some valuation factors and see if they can improve performance on the stocks we are trading.

These tests include survivorship bias, dividends etc and all of the other things required for an accurate test by the forum police..

These tests divide the stock universe, in this case all stocks listed in the United States that have fundamental data into 10 buckets. Each stocks receives an equal weighting in the tests.

In the first test we'll look at Price to Sale ratios and see if lower price to sales ratios generate alpha.

Pr2SalesTTM
Price to Sales Ratio, TTM
Category
RATIOS & STATISTICS / VALUATION
Ranking Default
When this factor is used in a ranking system stocks with lower values will get higher ranks
Full Description

This is the current Price divided by the Sales Per Share for the trailing twelve months. If there is a preliminary earnings announcement for a quarter that has recently ended, the revenue (sales) values from this announcement will be used in calculating the trailing twelve month revenue per share.

NOTE: Most Banks and Finance companies do not report revenues when they announce their preliminary quarterly financial results in the press. When this happens, the trailing twelve month values will not be available (NA) until the complete quarter is released.



Performance Graphs for each bucket.



The conclusion with this single factor ranking system show stocks with lower price to sales ratios tend to perform better than stocks with higher price to sales ratios. Note the testing reweights the bucket every 4 weeks.

Later on we'll be using these ranking system for stock selection in the stock screens we'll look at and or/portfolios we create..

In the next post we'll look at Price to Projected Earnings.

 

[zipzap]

This is another single factor ranking system which looks at Projected PE ratio.

The first graph below ranks the stocks in the universe into 5 groups, from highest PE ratio to lowest PE ratio and then breaks the list into 5 baskets showing the average return for each.



Again the basket showing the highest return comes from those stocks with the lowest projected PE ratio with an average annual return around 15% compared with the S&P500 at circa 3%. Below is the performance graph for each basket.



Next we'll look at some multi-factor ranking systems to see if returns increase from looking at more than one factor.

An example might be a valuation factor like PE ratio combined with an earnings surprise or even perhaps a growth forecast.

 

[sinner]

Sinner

Thanks for the lnfo/limks however the turnkey link just goes to a login screen.

Free signup link is directly below the login button.

RRRrrrrRRRRrrrrRRRRrrr, the self appointed forum police is back on the case. Sinner, I would suggest you have an in-depth look at Portfolio123 so you can provide in informed opinion, it' a very powerful program.

If you feel like anything I'm stating is incorrect or nonfactual in any way, feel free to report it to the actual forum police so they can take a look. For the record, my opinion is very informed, you won't catch me posting on this forum unless it's to learn something or to discuss something on which I have an informed opinion.

In the next group of posts I am going to demonstrate how fundamental data can be used for isolating a group of stocks to trade based upon a combination of fundamental and technical considerations.

The first step is to build a ranking system

For this post we'll have a look at some valuation factors and see if they can improve performance on the stocks we are trading.

These tests include survivorship bias, dividends etc and all of the other things required for an accurate test by the forum police..

So how much of this portfolio alpha is due to the "Monday effect" and "equal weight" phenomenons and actually have nothing to do with P/S ratio (which is easily the most industry specific - and therefore least robust across a broad basket - of the value factor ratios)?

Have you even bothered to check? Seems like a lot of Monday effect to me. Seems like you are fooling yourself on multiple levels here.

 

[sinner]

From what I can see in the chart, the return from end of '99 until end of 2011 is approx ~450%.

Here is the return from the equal weight CRSP portfolio over the same duration. As you can see the R^2 is very high and return is only ~100% less (depending on the day you pick).

So how much of your portfolio returns are actually do to picking the cheapest P/S quintile? Only approximately a quarter of the total returns, at best, if you're completely discounting the Monday effect.



EDIT: This portfolio is actually rebalanced annually, so my guess is if you rebalanced it monthly it would probably return closer to the equity curve as shown by the OP. So is the OPs alpha really derived from fundamental quantitative factors, or (much more likely) simply because he hasn't realised that equal weighting positions tends to provide a "free diversification lunch".

 

[zipzap]

Speeding up the process... In the table below I have run a test which includes a 2 factor ranking system combining a Valuation factor with a Growth Factor (called Sustainable Growth%) and run an optimisation which looks at each valuation factor individually with the Growth factor to see what valuation factor combines well with growth.

The value factors are shown below

PEExclXorTTM (Price to Earning Excluding Extraordinary items current)
Pr2FrCashFlTTM (Price to Free Cash Flow , Current)
EarnYield (Earning Yield)
PEG (PE Growth Ratio)
Prc2SalesIncDebt (Price to Sales Including debt current)
Pr2TanBkQ (Price to Tangible Book Value Current)
Pr2BookQ (Price to Book Current)
Pr2SalesTTM (Price to Sales)

The Growth Factor

Sustainable Growth %



Looking at the tables the Earnings Yield combined with Sustainable Growth % worked well with the best percentile showing a 17% return. the price to sales combined with growth seemed to work the least well, but the numbers still look reasonable.

 

[zipzap]

Interesting stuff Sinner, was going to mention that generally an equally weighted benchmark outperforms a capital weighted bench mark and that shows through in a lot of the screens, especially in those tests where all percentiles outperform the benchmark for that very reason..

Anyway will carry on with these self deluding tests, please excuse my ignorance.

 

[craft]

ZIPZAP

Is the historical raw data that the tests are running off visible/accessible to subscribers?

 

[sinner]

Interesting stuff Sinner, was going to mention that generally an equally weighted benchmark outperforms a capital weighted bench mark and that shows through in a lot of the screens, especially in those tests where all percentiles outperform the benchmark for that very reason..

Anyway will carry on with these self deluding tests, please excuse my ignorance.

Wait, so you actually understand what I'm describing, but you specifically didn't mention it?

Why not?

Seems awful misleading to me.

 

[McLovin]

ZIPZAP

Is the historical raw data that the tests are running off visible/accessible to subscribers?

There's a video on the website of the system and it appears no.

 

[zipzap]

Wait, so you actually understand what I'm describing, but you specifically didn't mention it?

Why not?

Seems awful misleading to me.

Sinner, you're clutching at straws...

How do you explain the increased returns as we work up the percentiles?

Also if you were to choose a valuation factor, which one/s would you choose and how would you compute the number and how would you adjust it?

Craft, you cannot export the data, however you can use the data with a program called Quantshare and maybe you can export it through that, not sure.

 

[sinner]

Sinner, you're clutching at straws...

How do you explain the increased returns as we work up the percentiles?

LOL very simply as a poor use of arithmetic averages to hide the facts?

If you take even a cursory glance at the P/S equity curve you can plainly see that the outperformance is not systematic across quantiles and as a matter of fact the "cheapest" quantile underperformed the second cheapest quantile until 2008 when the FASB Mark to Market rules were suspended and replaced with Mark to Unicorn. So it is very very easy to verify my statement that P/S is not robust valuation when you are working across multiple industry sectors because you can see it right there in the chart.

So, by the same token, OP can you please explain for those who don't know, why the return for the second cheapest P/S quantile was higher than the cheapest quantile until 2008? Having read the Shiller paper where he uses momentum to discard "value trap" sectors and the work of Eric Falkenstein on his defProb default model, I understand pretty thoroughly exactly why this is the case. Do you?

The P/E quantiles (even though you didn't include equity curves for all) are generally explained by a behavioural bias rather than actual fundamental pricing matter, and that behavorial bias is actually one you are falling for by suggesting a "Growth Factor" screen should be added in.

Also if you were to choose a valuation factor, which one/s would you choose and how would you compute the number and how would you adjust it?

It absolutely depends on the market, industry and index in question. At the macro level I am using CAPE balanced equally with MV/GDP, and "future 10Y returns" to build a least squares linear regression for forecast purposes (similar to as seen on alpha.turnkeyanalyst.com except I made my own version with multiple inputs not just CAPE), at the single name level I am using EBIT/TEV as defined in "Quantitative Value".

 

[sinner]

I have also found this paper to be inspiring but done no specific work to implement or test (yet):

papers.ssrn.com/sol3/papers.cfm?abstract_id=2051101

 

[zipzap]

This is all very academic however at least you are prepared to provide some valuable contributions which is good and probably adds value to this thread. I am going to try to work through your comments.

1. Survivorship bias - the data being used (S&P Compustat Point in Time database) includes survivorship bias so I hope that is sufficient. More detail can be found in the message board for the product on this issue.

2. Equal weighted portfolios show a higher return than capital weighted indices in general so this needs to be considered in results. However I might add it odesn't really matter, we are scanning for stocks that will hopefully generate a reaosnable equity curve so as long as we get there, that is more important than being to academic about it. Still it is important to isolate where the added returns are coming from. However difficult to quantify with the ranking system, one way it could possiblt be done is to run another ranking system which only includes those two items although you'd still need something to rank the stocks in some sort of order rather than looking at everything in aggregate... For me, if each quartile gets stronger with the particular ranking system being used it could possibly indicate some sort of bias.

3. When benchmarking need to compare apples with apples, one problem onwith the ranking system is that it will only benchmark against the S&P500 index and I am testing on all US stocks with adequate fundamental data available for testing. Once I get to using the ranking system in a stockscreen or Portfolio simulation any benchmark can be used but from a strictly acedamic standpoint (and I'm no academic... so I'm not going to win debates on that front) this probably isn't ideal.

4. Monday effect... not sure how this plays a part when the ranking system assumes 100% invested at all times and when re-weighting stocks are sold and purchased at the same time (Mondays open). The index itself should also reflect a Monday bias if that is the case... I might add that I am aware of some of these effects, many of my futures systems take advantage of end of month rallies, first trading day of the month effect and other seasonal biases. I never really found the Monday effect to be particularly strong, but must admit my Dax system favours Monday and Tuesday over other days for getting long.. anyway off topic...

5. Small cap stocks have outperformed larger cap stocks over time so that probably account for some additional outperformce in the ranking system. Leter when we benchmark against the right index through the Portfolio simulator or stock screen back test we can address this.

6. When comparing the tests I am running here and say the academic ones mentioned, bear in mind that the data sets being tested on might be different so I daresay that will account for some of the differences in results, other than my ignorance.


I want to make the comment that in order to be a trader or investor it's about generating profitable ideas and you don;t need to be an academic to do that. Yes, supporting the idea with a premise is very important.

I might also mention that P123 is used in some United States Universities and has a strong academic following, the problem here, from Sinners persepctive is more likely me and my use of it rather than the program itself... Sinner I would suggest you actually look into the program first, are you attempting to discredit me or more specifically my use of the program or the program itself? Let's get clear on that.

Sinner, I'll move on to your comments in your previous post in my next post.. got to go back and see what they are.. I actually don;t think we are that far apart, in fact you're more knowledgeable than me with this type of analysis so keep posting, it's making me step up and go back and check things as well.

 

[zipzap]

LOL very simply as a poor use of arithmetic averages to hide the facts?

I'm not sure I agree with this, it's a very convenient comment so please ellaborate.

From my perspective, the testing is simple enough, rank all stocks from 1-about 1500 (there are about 1500 stocks in the universe being tested) based upon the valuation criteria and then break it into five baskets are re-weight the basket once per month. The baskets showing the lowest value criteria showed steadily increase outperformance as we moved along the percentiles. How can you explain this in terms of Monday effect and equal weighting effect when that should show an equal weight for each basket. I might actually run this test and we can use that return as the benchmark rather than the index.

If you take even a cursory glance at the P/S equity curve you can plainly see that the outperformance is not systematic across quantiles and as a matter of fact the "cheapest" quantile underperformed the second cheapest quantile until 2008 when the FASB Mark to Market rules were suspended and replaced with Mark to Unicorn. So it is very very easy to verify my statement that P/S is not robust valuation when you are working across multiple industry sectors because you can see it right there in the chart.

That's fine and thanks for the info, I was illustrating the program, the test also suggested other valuation criteria showed higher returns.

So, by the same token, OP can you please explain for those who don't know, why the return for the second cheapest P/S quantile was higher than the cheapest quantile until 2008?

Maybe the testing was run on different data, I certianly did not adjust for moment, although this can be done. The results were run from 99-Now so perhaps if the test was just run through to 2008 it would show similar results to those you've mentioned in your next comment. Seriously..do you want me to run the same test but only to 2008..I can do this.

Having read the Shiller paper where he uses momentum to discard "value trap" sectors and the work of Eric Falkenstein on his defProb default model, I understand pretty thoroughly exactly why this is the case. Do you?

I have not read the Shiller paper nor Eric Falkenstein so your understanding is obviously far better than mine. In terms of avoiding the value trap this could be accomplished by creating a multifactor model such as shown below. You would actually be a pretty expert user of the softaware straight away with your undertstnading of fundamental stock screening and quant analysis as you are already aware of some of the pitfalls etc and know what to look for already.



And the parameters tested.

The P/E quantiles (even though you didn't include equity curves for all) are generally explained by a behavioural bias rather than actual fundamental pricing matter, and that behavorial bias is actually one you are falling for by suggesting a "Growth Factor" screen should be added in.

Maybe, however my spproach is to test any number of combinations to see if they add value, if it doesn't discard it and move on. Your backghround research certainly would help in building a good ranking system and. This also helps me form a view on the different aspects that combines well together..the downside it is is very time consuiming. At this early testing stage this was more about showing some of the things we can look at as opposed to falling for a value trap.

It absolutely depends on the market, industry and index in question. At the macro level I am using CAPE balanced equally with MV/GDP, and "future 10Y returns" to build a least squares linear regression for forecast purposes (similar to as seen on alpha.turnkeyanalyst.com except I made my own version with multiple inputs not just CAPE), at the single name level I am using EBIT/TEV as defined in "Quantitative Value".

Thanks for this, it appears a lot of what you mention is covered from the site you've provided links for.. We can test EV/EBIT or EBIT/TEV etc but it's difficult (but possible, just takes time to get it right) to build a ranking system on these but it requires a multifactor approach and the elimination of some of the companies (with an EV/EBIT <say 3 or above say 10 (EV/EBIT)) which distorts the ranking system. Attached is a default one that include EV/EBIT.



Graph

 

[craft]

I might also mention that P123 is used in some United States Universities and has a strong academic following, the problem here, from Sinners persepctive is more likely me and my use of it rather than the program itself... Sinner I would suggest you actually look into the program first, are you attempting to discredit me or more specifically my use of the program or the program itself? Let's get clear on that.

I’ve had a bit more of a look trying to work out the data visibility question and although the data is not visible the software looks like a pretty impressive bit of kit.

Whether a fundamental system developed on historical data is robust going forward has all the same issues as price driven systems and I’m staying out of that debate.

However as a fundamental data back tester and a scanner for candidates to investigate further goes, it looks pretty good. (Damn site better and more transparent then another piece of junk being flogged in AUS)

ZZ is every stock on the ASX included in the database?

 

[zipzap]

Hi Craft, at this stage it is only for US equities. Adding ASX would be a 4-6 month project and costs $$ but well worth it in my opinion.. however to answer your question it only does US stocks at the moment.

 

[nulla nulla]

Rightio, so let's have a look at stock selection. To do this I use Portfolio123 so we're going to focus on US shares. www.portfolio123.com.au . Btw I am getting ASX and NZX shares added to Portfolio123 so stay tuned.

Technical System on ASX Property Trusts. From 2005 to now.

View attachment 54873

Hi Craft, at this stage it is only for US equities. Adding ASX would be a 4-6 month project and costs $$ but well worth it in my opinion.. however to answer your question it only does US stocks at the moment.

okay i'm getting confused here... In your early posts, you are in the process of adding the ASX and NZX shares, then you run a test on the REITs' for me (thanks) then it is "only US equities".

As suggested I emailed you about getting the output of the actual trades making up the earlier summary report, in excel format rather than txt, so I could assess the individual shares involved. I felt this would be beneficial in the share selection process.

Are you able to provide an excel output of the REIT trades referred to in your summary?

thanks & regards

nulla

 

[zipzap]

okay i'm getting confused here... In your early posts, you are in the process of adding the ASX and NZX shares, then you run a test on the REITs' for me (thanks) then it is "only US equities".

As suggested I emailed you about getting the output of the actual trades making up the earlier summary report, in excel format rather than txt, so I could assess the individual shares involved. I felt this would be beneficial in the share selection process.

Are you able to provide an excel output of the REIT trades referred to in your summary?

thanks & regards

nulla

1. Nulla, the screening software/portfoliobuilder P123 only does US equities, the technical overlay I programmed in Trade Navigator so can be applied to any stocks if you have the data.

Draw the dinstinction between the program we are using to select the stocks to trade (P123) (only does US equities)

The software I programmed the technical system into (Trade Navigator) (Has ASX data) which would be used for entering and exiting the stocks chosen above using the technical rules I spelt out.

2. You can open a .txt file in Excel ...so not sure what the problem is.

Cheers
ZZ...

 

[nulla nulla]

1. Nulla... You can open a .txt file in Excel ...so not sure what the problem is.

Cheers
ZZ...

Okay, email me the .txt file then when you get a minute.

cheers