GuruFocus Premium Membership

Serving Intelligent Investors since 2004. Only 96 cents a day.

Free Trial

Free 7-day Trial
All Articles and Columns »

GuruFocus Research: What worked in the market from 1998-2008? Part I: Introduction of Predictability Rank

September 30, 2008
insider

gurufocus

353 followers
With the stock market in turmoil, even a lot of our Gurus have suffered great losses. GuruFocus recently conducted a back test study of Warren Buffett’s strategy of “buying good companies at fair prices” for the years from 1998-2008. This is the first of a series of reports on the study results.

Since Oct. 2007, the financial market has been hit by a crisis to a degree unmet in many decades. Strategies that worked for many years see their first “black swan”, and many of our gurus have also suffered permanent losses of capital. Things that worked for many years just stopped working. At the time of this writing, Lehman Brothers filed bankruptcy and Merrill Lynch was sold to Bank of America in a fire sale. A week ago, Fannie Mae and Freddie Mac were taken over by the U.S. government. With the bear market in full swing, all these things are now happening on weekends.

At the time of crisis and uncertainty, any defects in your investing strategies that have been hidden for years have surfaced. If a strategy can pass Warren Buffett’s Rule #1 test in the recent market , it’s probably more robust than most others.

With this in mind, we did a back testing for the performances of Warren Buffett strategy from 1998 to 2008. We recently introduced a new feature:10-year financials, and this data has been used for our back testing.

This is the Part I of the study. Also

Part II: What worked in the market from 1998-2008? Part II: Role of Valuations
Part III: Intrinsic Value, Discounted Cash Flow and Margin of Safety

The performances from 1999 to present are monitored with real portfolios. There are here: http://www.gurufocus.com/stock_ideas.php

So what is Warren Buffett strategy?


Warren Buffett said many times that the companies he likes are:


1. Simple businesses that he understands
2. that have predictable and proven earnings and
3. with economic moat
4. those can be bought at a reasonable price.

It is hard to quantify “simple businesses that he (Buffett) understands”, so we will focus on the other three characteristics instead. As we will later show, the businesses that have predictable and proven earnings are usually also simple businesses that an average person could understand.

Predictability of Businesses


In our database there are 2403 stocks that have been traded from Jan. 2, 1998 to Aug. 31, 2008. We have the complete 10-year financial data and trading prices of these companies for this period. We rank the predictability of these companies based on the consistency of their revenue per share and EBITDA (earning before interest, tax, depreciation and amortization) per share over the past ten fiscal years, and study the correlation between the stock performances and the predictability of the business.

We use 10-year financial data because we think that 10 years is enough to cover a full cycle for most businesses. 10 years should also be enough for the business value to be reflected on the stock prices. The recent financial crisis is perfect for this study because it makes the business cycle more complete.

Fig. 1 shows the revenue per share, EBITDA per shares of Walgreen (WAG), Wells Fargo Bank (WFC), Apple Computers (AAPL), and USG Corp. (USG). When comparing the revenue and EBITDA performances of the predictabilities of their business becomes quite apparent.






As we can clearly see, Walgreen has far better business predictability than Wells Fargo, which has much better business predictability when compared to Apple Computers (APPL) and USG Corp (USG). We ranked the predictability of the 2403 companies. Any companies that ever had operating loss in any fiscal year during the past 10 years are considered unpredictable. Therefore, among the four companies mentioned above, Apple Computers and USG Corp are unpredictable.

Among the 2403 companies, there were only 570 that are predictable according to our definition. As we will show below, these companies had much better stock performance than the others, and more importantly, the chance of losing money when investing in these is much smaller.

Study Assumptions and Biases


A few things to clarify before we report the correlation between the business predictability and the investment returns. Our study may be subject to these biases and assumptions:
  • Dividend yields are not counted for investment returns
  • Effects of price changes due to spin-offs may not be fully adjusted
  • Study is subjected to survivorship bias due to de-listing, bankruptcy, LBO, M&A, etc.

Dividends are certainly an important part of investment returns. However, in this study we ignore the dividend returns. This will reduce the investment returns by at about 2% a year on average. Since predictable companies tend to pay higher dividends and pay them more regularly, this bias favors the average return of non-predictable companies.

We ignore the price adjustment due to spin-offs during the past 10-year for these 2403 companies. Although we do not have the accurate number,we believe the error caused by this factor is small.

Survivorship bias is another important factor in this study. In 1998 there were certainly much more than 2403 stocks traded in the exchanges. However, only about 2403 survived until the present day. 1998 was a year of IPO and dotcom bubble; many of those stocks were later de-listed. We believe that survivorship bias strongly favors non-predictable companies because they tend to lose more money and become insolvent.

The correlation between business predictability and investment returns
  Predictables Non-Predictable All stocks
Total Number of stocks 570 1833 2403
Total lost money 61 830 891
Total lost more than 50% 18 412 430
Total lost more than 90% 4 86 90
Average gain 260.6% 100.0% 138.1%
Median gain 150.0% 13.0% 39.0%
Maximum Gain 2852.0% 11483.0% 11483.0%
Maximum Loss -100.0% -100.0% -100.0%
Annualized Average gain 12.7% 6.7% 8.4%
Annualized Median gain 8.9% 1.1% 3.1%

The correlation between the business predictability and investment return of a company is shown in Table 1. The table header “Top 100” means the top 100 ranked predictable companies. “Second 100” means the next 100 top ranked predictable companies, and so on. “Non-predictable” means that the company had at least one loss year during the past 10 years, so the business is non-predictable to us.

All 2403 stocks:


As shown in the last column of the table, among the 2403 stocks studied, 891 of them, or about 37% are still losing money after a 10-year and 8-month holding period. About half of these stocks lost more than 50%, and 90 of them lost more than 90%. Therefore, we can conclude that the, “buy and hold” strategy does not work if you buy bad companies.

The 2403 stocks show an average gain of 138% over the 10-year and 8-month period and a median gain of 39%. In annualized terms, these stocks had an annualized average gain of 8.4%, and an annualized median gain of 3.1%. During the same period S&P500 had annualized average gain of 2.7%. All numbers do not include dividends.

The average annualized gain of 8.4% of these 2403 stocks is much higher than the 2.7% of S&P500. This can be caused by the non-weighted average of these stocks. Also, the survivorship bias mentioned above can have a large contribution to the excess gain. Since in 1998, a lot dotcoms disappeared, even large companies such as Enron or WorldCom were de-listed. If all these de-listed and bankrupt companies were counted in our research, the average and median gain would be greatly reduced.

The median gain of 3.1% here is close to the annualized average gain of S&P500 during this period. We will use this number as a reference for the returns of each group.

The largest gainer for the companies was Hansen Natural Corp. (HANS). It gained more than 114 times during the period. Hansen is engaged in the business of marketing, selling and distributing so-called ”alternative“ beverage category such as natural sodas, fruit juices and juice cocktails. The top 10 gainers during the 10 years are shown in Table 2:

 

All Non-Predictable Companies


There are 1833 non-predictable companies, and every one of them had at least one year of operating loss. As shown in Table 1, the annualized average gain of these stocks is 6.7%, and the annualized median gain is 1.1%. All these are lower than the averages of the 2403 stocks. Among all the 1833 non-predictable companies, 830, or 45% of them are still in loss after a holding period of 10 years and 8 months. About 50% of them lost more than 50%, and 10% of them lost more than 90%.

All Predictable Companies


There are 570 predictable companies. The annualized average gain of these stocks shows a much higher 12.7% when compared to the non-predictable companies, and the annualized median gain is 8.9%. These numbers are better than the average of all stocks by more than 6% a year.

The possibility of loss is also dramatically lower among the predictable companies. Among the 570 predictable companies, 11% are in loss with the 10-year and 8-month holding period, which is much smaller than the average of 37% for the non-predictable companies. Predictable companies that lost more than 50% are reduced to 30% of those losers.

Introduction of Predictability Rank


For the 570 predictable companies, we have seen strong correlation between the predictability of businesses and the stock performances over the past 10 years, regardless of the valuation of business at 1998. Accordingly, we have ranked the business predictability from 5-star to 1-star, as shown in this table.
Predictability Rank 5-Star 4.5-Star 4-Star 3.5-Star 3-Star 2.5-Star 2-Star 1-Star (non-predictable) Average among all
% out of all 2403 stocks 3.3% 2.9% 3.7% 3.3% 3.3% 3.7% 3.3% 76.3% 100%
% that are in loss (10y) 3% 10% 8% 9% 11% 18% 16% 45% 37%
Average gain (10y) 364.6% 330.9% 278.0% 235.1% 243.5% 227.8% 154.8% 100.0% 138.1%
Median gain (10y) 238.5% 193.5% 171.0% 159.0% 132.5% 113.5% 87.0% 13.0% 39.0%
Maximum Gainer 2228.0% 2547.0% 2452.0% 2852.0% 2432.0% 1892.0% 1807.0% 11483.0% 11483.0%
Max Loser -82.0% -53.0% -67.0% -100.0% -83.0% -100.0% -78.0% -100.0% -100.0%
Annualized Average Gain 15.4% 14.6% 13.2% 12.0% 12.2% 11.7% 9.1% 6.7% 8.4%
Annualized Median Gain 12.1% 10.6% 9.8% 9.3% 8.2% 7.3% 6.0% 1.1% 3.1%


5-Star Predictability has a typical business performance chart like that of Walgreen's:




Only 3.3% out of the 2403 stocks are ranked 5-Star. In the back testing for 1998-2008, only 3% of the 5-Star stocks are still in loss. Their annualized median gain was 12.1%, which is 9% a year above the average of all stocks.

4.5-Star Predictability: Only 2.9% out of the 2403 stocks are ranked 4.5-Star. In the back testing for 1998-2008, 10% of the 4.5-Star stocks are still in loss. Their annualized median gain is 10.6%, which was 7.5% a year above the average of all stocks.

4-Star Predictability has a typical business performance chart like this.


 

Only 3.7% out of the 2403 stocks are ranked 4-Star. In the back testing for 1998-2008, 8% of the 4-Star stocks are still in loss. Their annualized median gain is 9.8%, which was 6.7% a year above the average of all stocks.

3.5-Star Predictability: Only 3.3% out of the 2403 stocks are ranked 3.5-Star. In the back testing for 1998-2008, 9% of the 3.5-Star stocks are still in loss. Their annualized median gain is 9.3%, which was 6.2% a year above the average of all stocks.

3-Star Predictability has a typical business performance chart like this.


 

Only 3.3% out of the 2403 stocks are ranked 3-Star. In the back testing for 1998-2008, 11% of the 3-Star stocks are still in loss. Their annualized median gain is 8.2%, which was 5.1% a year above the average of all stocks.

2.5-Star Predictability: Only 3.3% out of the 2403 stocks are ranked 2.5-Star. In the back testing for 1998-2008, 18% of the 2.5-Star stocks are still in loss. Their annualized median gain is 7.3%, which was 5.2% a year above the average of all stocks.

2-Star Predictability has a typical business performance chart like this.


 

Only 3.3% out of the 2403 stocks are ranked 2-Star. In the back testing for 1998-2008, 16% of the 2-Star stocks are still in loss. Their annualized median gain is 6%, which was 2.9% a year above the average of all stocks.

1-Star Predictability has a typical business performance chart like this.


 

76.3% out of the 2403 stocks are ranked 1-Star, these companies are not predictable, they had at least one year of operating loss over the past 10 years. In the back testing for 1998-2008, 45% of the 1-Star stocks are still in loss. Their annualized median gain is 1.1%, which was 2% a year below the average of all stocks.

Not-Ranked: There are more than 10,000 stocks trading on the major exchanges. We only ranked 2403 of them; all others are not ranked. They are not ranked mainly because most of them do not have more than 10 years of history, or we do not have 10-year financial data if they do.


The median gains and percentage of losers at different predictability ranks are shown in the charts below. We can clearly see that regardless of stock valuation, companies with higher Predictability Rank had much better stock performances. The possibility of losing money with long term holding period is smaller, too.






Predictability Watch


All businesses have challenges. Competitions, economic cycles and business execution may make businesses deviate from its previous tracks, and business predictability may change. In order to highlight this, we create a feature called Predictability Watch, which means that a predictable business shows sign of deviating from previous predictability. This deviation can be temporary; it can also be permanent.

The star-rating with a red frame shows that the company is on Predictability Watch.

We update the Predictabilities of Business quarterly, as companies report their financials once a quarter.

How Not to Lose Money?


Our study proves that, just as Warren Buffett said to us many times, by buying business with predictable and proven earnings you will have a much smaller chance of losing money.

Go to http://www.gurufocus.com/predictable.php for the current list of top-ranked Predictable Companies. This list is for Premium Members only. If you are not a Premium Member, you are invited for a Free Trial.

What is the next step?


So far we have not discussed the valuation and economic moat of the companies. In the following reports we will show our results regarding the roles of stock valuation and economic moat played in the past 10 years, as we discovered in the back testing study.

We also created a young Buffett Screener based on the 10-year back testing study and GuruFocus’s understanding of Warren Buffett investing, which covers the aspects of business predictability, economic moat, and valuations. We did a back test on the performances of top 25 five stocks in young Buffett Screener for the periods from Jan. 2005-Aug. 2008. We never had a losing year, and the results outperform the market by great margins. The results will be shown in the following report.

We opened a new menu item called “Research & Strategies”. The rank of business predictability is listed under this menu. The current Buffett Screener is also listed there. Go to http://www.gurufocus.com/predictable.php for the current list of top-ranked predictable companies.

Take a Free Trial of GuruFocus Premium Membership.

Also read:
Part II: What worked in the market from 1998-2008? Part II: Role of Valuations
Part III: Intrinsic Value, Discounted Cash Flow and Margin of Safety

The performances from 1999 to present are monitored with real portfolios. There are here: http://www.gurufocus.com/stock_ideas.php

 

About the author:

gurufocus
GuruFocus - Stock Picks and Market Insight of Gurus

Rating: 3.1/5 (173 votes)

Voters:

Comments

mike shearin
Mike shearin - 6 years ago
A very good article. I would like to see the returns based on risk per dollar invested. If the risk is lower, the returns are usually lower as well. I do know that Buffett's #1 goal is to not lose money and his #2 rule is to never forget rule #1. This seems to correlate accordingly.
saturnc15
Saturnc15 - 6 years ago
Dividends are not counted on returns? That makes no sense.
gurufocus
Gurufocus premium member - 6 years ago
As we discussed, dividends are an very important part of investment returns. However, not considering dividends does not change the conclusion that more predictable companies have best stock performances.

bakerkn
Bakerkn - 6 years ago
Your back testing is assuming time-travel is possible. You are using the predictability over the last 10 years to decide what stocks to have bought 10 years ago.

10 years ago you would not have had the 1998 to 2008 information at hand. So for this study to have validity, you need to show that stocks chosen based on historic predictability data have outperformed.

For example, stocks chosen in 1998 based on 1988 to 1998 data. Or, if you do not have data going back that far, stocks chosen in 2003 based in 1998 to 2002 data.
gurufocus
Gurufocus premium member - 6 years ago
What you said is certainly valid, just as past performance is not a good indicator of future performance.

But on the other hand, past Underperformance is a good indicator of future Underperformance. If a business did poorly at bad economic times before, the chances are it will do poorly again in future bad economic times.

Business predictability is mainly determined by the nature of the business, less dependent on who runs it.

In order to further validate the idea, we used the data before Jan. 2005 to test the performances of stocks in 2005, and data before Jan. 2006 to test those for 2006, and so on. In this way, time does not travel back. The results will be reported in the next article. By the way, the results are similar.


xjz27
Xjz27 - 6 years ago
Wonder how this is compared with S&P's quality rank? Did you do a study on this?
gurufocus
Gurufocus premium member - 6 years ago
No, we did not study the predictability of S&P500 companies. But this is certainly an interesting idea, and we will put this as part of our study.

Thanks!
gerraldchew
Gerraldchew - 6 years ago
Pretty good article...thumbs up!!!

dani
Dani - 6 years ago
so???

what are the predictabil companies???

were can i get the list????
acedomaine1
Acedomaine1 - 6 years ago
Y'all note that Bakerin is getting at the heart of the study which is probalistic based as it depends on averaging concepts and thus generalizing the data. The upshot is that this sort of analysis wouldn't be acceptable in Finance 101 because it doesn't show all the data and is making an empirical inference that intuitively says thata glance at the data makes the whole matter obvious. If we had the actual data, the antithetical reply of the cautious sceptic would consist in pointing out the wide dispersion in the results and the relatively high value of some of the samples; as a consequence I would refuse to make any inference from the data.Is the conclusion based on "predictability" sound? Is the reasoning cogent? Could not such a result occur by chance even when no real differences exist? Are the empiral data too widely disperrsed to make any sensible inference? And how do we know whether the inference based on predictability is correct? It is the avowed purpose of statistical and financial analysis to avoid the confusion to which your intuitive research leads : how are we to know whether the observations of the data really satisfy the presuppositions?What good does it do to define so carefully the criteria of "predictability" , when such a criteria are only valid provided doubtful conditions hold? And, finally, is it fair to leave the matter up to the "common sense " of the reader, and hope that everyone will have some sort of "mutual understanding" on this matter of "predictability"? The history of financial analysis provides a cumulation of evidence against any view that all folks mean the same thing by "predictability" and its corollary "truth."
batbeer2
Batbeer2 premium member - 6 years ago
I find this somewhat hard to read.

As i see it, the fact that a company has increased revenue and earnings in a linear fashion - or more accurately in a manner that is easily described with a first order function - in the past does not guarantee it will continue to do so from now on.

However.....

A company that has increased revenue and earnings in an erratic fashion in the past is VERY unlikely to be predictable from now on. Ergo we eliminate all the unpredictable companies. We might not eliminate ALL unpredicable companies but we have eliminated most.

Case in point...

Who is willing to bet with 25% accuracy what the earnings for JNJ or Wallgreen will be in 2010.

Who is willing to bet with 25% accuracy what the earnings for USG or BVF will be in 2010.

I'm willing to bet there are more poeple in the first group than in the second. A lot of poeple will be in the first group for the wrong reasons but that is another matter.
toroariza
Toroariza - 6 years ago
just one word: Amazing

softwareNerd
SoftwareNerd - 6 years ago
That's some great work. Well done. Thank you.
redcorolla95
Redcorolla95 - 5 years ago
How is predictability measured? Are we looking at standard deviation of growth rate of EPS?
gurufocus
Gurufocus premium member - 5 years ago
Predictability Rank measure the consistency (including standard deviations) of both the growth of revenue and operating earnings. The Buffett-Munger screener filters the companies that are highly predictable and have competitive advantages, and no excess debt.

BBH
BBH - 5 years ago
According to the predictability ranking, Warren Buffett's recent buys COP, ENT, IR are unpredictable companies (ranked 1 ). Does it mean factor #1 (Simple businesses that he understands) outweighs rest of the factors. If that's the case then what's the use of predictability ranking ?

Am I missing something here?

wiglebot
Wiglebot - 5 years ago


So now we need to know how to predict the revenue.

This is not showing correlation or predicting anything, but it is a good study to represent correspondence.

This study shows that the data in the database has a few factors that create a sub set that shows a consistent market price raise. But:

--How many factors are in the data?

--How many other factors can be combined to give better results?

--Does the combinations of business attributes outperform the combination market attributes?

--How many companies would be on the "predictable" list for each year?

--How many companies drop off each year?

I think what you show is exactly how people value a stock, not what stocks will increase in price in the future. The earning raises steady, the revenue raises steady and the price reflexes that. The revenue tanks, so does the price. But that puts you back to the drawing board because now earnings and revenue need to be predicted for this to mean anything.






mreidmd1
Mreidmd1 - 5 years ago
Is the annual growth trendline based upon a least squares regression calculation?

Paleface
Paleface premium member - 3 years ago
Guru Focus has worked with rare sincerity on developing valid value investing screening parameters. I am very impressed. The results are good enough to warrant further study! The Predictability screener is the foundation of the Buffett-Munger and UPC screeners, which potentially produce stellar results but presumably are too difficult to backtest thoroughly. (?) I also consider Predictability to be like a floating list of Index or blue chip stocks from which to choose minimal-risk investments using any other parameters. Therefore I would suggest extensive backtesting of Predictability so as to nail-down the ups-and-downs of Predictability as a fundamental value in every market. My humble suggestions...

  1. Let's please have Predictability backtests for our last 4 unique decades? The stagflating 1970's, the steady-growth 1980's, the tech-booming 1990's and the fund-busting 2000's? Also ideally, all results should be based on holding all picks for 1 year. Reevalute annually for any new predictibles and re-investing in 1-year old predictibles that are still predictible. (And of course, as already discussed here, being sure to use only "previous" data for picking at any time period.)
  2. Also for each decade's backtest, please include year-by-year results? So we also have a better idea of the relative volatility.
  3. By the way, I think "Dependability" would be a more accurate screen title than "Predictability." For one thing, you don't want to give the impression that you are claiming to "predict" the market. Also then you might develop short-selling screens based on "undependability." ("Unpredictability" would not sound quite right!)
PapaBear
PapaBear - 3 years ago
You should look into the literature on earnings persistence for the predictability of earnings. Sloan (1996) is a good starting point. AR(1) models can be used to find relative measures of predictability for example.

billbyte
Billbyte - 3 years ago
This is a great start but I do agree that we need to take various 10 year periods and "predict" the next 10 years for many 10 year periods to better understand the reliability of the model.

Once, there was a "Nifty fifty." In any 10 year period [and at an accelerating pace] there will be disruption of previous business models, managements become complacement and do stupid things that appear smart on the surface. Governments change the rules that made consistency possible. Moats can be filled and crossed.

The predictability watch is smart as is the evaluation of the company moat.

There are many "backtested" models that represent the effects of unconsidered variables. Many of these fail on a forward basis.

Net: Good work but we would be more comfortable with the process if it involved more than a single 10 year backtest.
UVInvestors
UVInvestors premium member - 2 years ago
What he says below is correct. This study is invalid without 1988-1997 data. I would suggest redoing the study using 5 years of data instead. You could use 1998-2002 data for coming up with a 5-star list, then look at performance numbers from 2003-2011. Using 5 years for reliability will still mimic Buffett's strategy. For example, he recently bought DTV and this company lost money 7-8 years ago but now is profitable.

"Your back testing is assuming time-travel is possible. You are using the predictability over the last 10 years to decide what stocks to have bought 10 years ago. 10 years ago you would not have had the 1998 to 2008 information at hand. "
jayb718
Jayb718 - 2 years ago
What about the companies whose operating cycle is predictable? Where do those rank?
Steve Kalos
Steve Kalos - 1 year ago
Why do the predictability ranks jump around frequently and, sometimes substantially? For example, AAPL's rank was somewhere between 4 and 5 several weeks ago, had fallen to 1 by the beginning of this week and is now 4.5.

gurufocus
Gurufocus premium member - 1 year ago
"Why do the predictability ranks jump around frequently?" They shouldn't and normally they don't. But we are in the process of changing the data source. That may cause some confusion.

This should be done by the middle of January, 2013.

UVInvestors
UVInvestors premium member - 1 year ago
What is the actual formula you used to rank eps predictability? Did you use historical EPS standard deviation %? i.e. (Std Dev/Mean), then sort them?
TonyPow
TonyPow premium member - 11 months ago
Great article.

If you played market timing, you could have avoided big losses as described in my blog. Since 2000, you have to exit and reenter the market only 3 times and the chart is quite easy to use.

http://tonyp4idea.blogspot.com/
TonyPow
TonyPow premium member - 11 months ago
I'm a new subscriber. From what I see so far, this subscription is fabulous and I will include it in several of my books (amazon). I'm especially praise the F-Score in your 'All-in-one Screen'.

I do not see you've a historical database to test the screens. However, you have a lot of performance data and for the price I do not complain. I came across the 10-year Fundamentals in this article. Would you explain how to use this info. Thanks for a great article again.
Oki
Oki - 3 months ago

Great site! Your predictability rank is based on the "consistency" of revenue per share and EBITDA per share over the past ten fiscal years. I would never invest in a formula that I don't fully understand so how are you defining "consistency" mathematically? Are you doing a linear regession? A Non-linear regression (and if so, what type)? Or are you simply comparing the coefficient of variation in each? If the latter then a company's revenue per share can be steadily falling and display a low std deviation. Are you really looking at consistency in revenue per share growth, for example? Please be more specific in your description otherwise I'm sure many potential users will be unwilling to put their hard earned money at risk. Thanks!

Please leave your comment:


Get WordPress Plugins for easy affiliate links on Stock Tickers and Guru Names | Earn affiliate commissions by embedding GuruFocus Charts
GuruFocus Affiliate Program: Earn up to $400 per referral. ( Learn More)
Free 7-day Trial
FEEDBACK