Profiles Of Good Investments & Statistical Computing

{ Euclidean Q1 2011 Letter }

Certain information from this letter relating to individual positions has been redacted

In our 2010 year-end letter, we talked about four specific investments to highlight the types of opportunities favored by our systematic approach to long-term investing.

Following the letter, we received some questions asking whether our approach was truly mechanical or whether we occasionally intervened in the decision making process.  In this letter, we thought we should begin with a clarification:  When we have cash to deploy, it goes into the highest-rated, available opportunities regardless of how we might feel about them as potential investments.  We oversee our models and review companies' financials to ensure data integrity, but we adhere to the process. 

This fact, when clarified, sometimes evokes other questions or comments such as:

  • “As you’re investing based solely on historic information, what do the best investments look like?”
  • “Ah! I can think of a company that would trick your models!”

In advance of a performance and operational update, we will share some thoughts relating to these points.

What Do The Best Investments Look Like?

As you know, our approach involved exploring the record of public company operating histories.  We hoped that, by digesting the quantifiable experience one could have obtained by investing in public companies over the past 40 years, we might find timeless answers to this question.

What did we find?  We found that the best (that is, the best we could identify) method for using historical financials to evaluate a company as a potential investment involves examining a handful of variables that characterize the following four concepts. 

  1. High and persistent returns on capital – Evidence that a company has, through some combination of good management and a strong business model, proven able to generate high returns on the money it has deployed in its business. 
  2. Long and consistent operating history – Confirmation that a company has operated for a long enough period and its results follow enough of a trend or cycle that there is some basis for understanding the character of its business. 
  3. Conservative balance sheet – Comfort that a company has not taken on excessive commitments given the scale of its business. 
  4. Inexpensively priced – Opportunity to purchase a company at a very low price in relation to its historic ability to generate cash.

We found that if you are evaluating the opportunity to buy a company based solely on its historic financials, you want all of those four concepts to be true.

“Ah! I can think of a company that would trick your models!”

Unfortunately, our approach is far from perfect.  It is susceptible to errors of both commission and omission.  Specifically:

  1. There are many instances of companies proving to be fruitful investments when one or more of those four key qualities are missing.  For example, highly leveraged firms – especially when they do not have to navigate difficult downturns – can deliver explosive returns.  Also, consistently innovative companies can be good investments even when they are offered at prices that might seem rich in relation to their historical financials.
  2. There are also instances when companies will prove to be poor investments even when all four key qualities are present.   For example, outside factors like new regulations, new technologies, or other evolving industry dynamics may cause the results of a historically strong company to degrade over time.  When this happens, one can lose money even with a historically strong company purchased at a low price.

It is this second point that is worth focusing on since we are judged primarily on the investments we make, not the ones (for better or worse) that we pass on. 

Many people wonder about Euclidean’s susceptibility to “value traps” and sometimes pose us with the challenge of, “Ah! I can think of a company that would trick your models!”  When this challenge is made, people most often ask if Euclidean’s models would have favored Enron back in 2000.  It is a good question because Euclidean focuses on historical financials.  As Enron’s fall from grace was precipitated by a massive fraud where their historical financial statements were misrepresented, it is reasonable to ask if our models would have been tricked in this instance.  

Although we admit that both machines and humans are susceptible to fraud, we can safely say Euclidean would not have invested in Enron.  The company would have been ruled out because it generated too low a return on the money it invested in its business and was offered only at extraordinarily high multiples of earnings.  

There is no question, however, that we can look silly from time to time.  In our year-end letter, we talked about our biggest loser of 2010, MEMC Electronic Materials.  MEMC looked like a great company offered at a low price.  It now looks, however, like MEMC might have a commodity business that was temporarily hiding in high-margin clothing.  We lost money with MEMC.   

If Euclidean had been in business for the past 30 years, we would have lost money on many other investments.  We would have likely invested in an airline on the dawn of deregulation of that industry.  Our models would have seen airlines as historically good businesses selling at low prices.  This would have proven unfortunate.  As their earning power eroded in the midst of newly unleashed price competition, airlines in general proved to be great destroyers of capital.

It is very difficult, though, to correctly identify these situations in advance.  There are countless examples where, because of some undue fear about the future, the market presented incredible investment opportunities.  Every successful investment we have profiled in prior letters was made possible by this dynamic.  

{redacted text }

So, when we receive the comment, “Ah! I can think of a company that would trick your models!” we reply that one probably can, but that it is easier to do in retrospect than in advance.  And, we encourage them to see how that challenge misses the point.  The important question is not whether our models can be wrong from time to time (they can and have been).  Rather, the right question, which we will explore in the next section, is: Can you look silly from time to time and still have an approach that does consistently well over long periods of time?  

Statistical Computing - Watson on Jeopardy

Earlier this year, IBM’s artificial intelligence system named “Watson” was showcased on the game show Jeopardy.  On the show, Watson played a two-game competitive match against the two greatest Jeopardy champions of all time – Ken Jennings, who once won 74 consecutive Jeopardy matches, and Brad Rutter, who is the show’s biggest all-time money winner. 

This was a real test of the evolution of computing and artificial intelligence.  Whereas artificially intelligent systems had made strides in the past, notably back in 1996 when IBM’s Deep Blue beat the world’s #1 player in a game of chess, there were real questions regarding whether a machine could succeed in a game, or any endeavor, characterized by metaphor, nuance, and context.

In the two-game Jeopardy match, there were several instances where Watson looked plain silly.  One time, after Ken Jennings had offered the wrong answer to a question, Watson subsequently buzzed in with the same wrong answer!  Another time, in Final Jeopardy, the category was US Cities. The clue: "Its largest airport is named for a World War II hero, its second largest for a World War II battle."  Watson, strangely, came up with the response: "What is Toronto?"   This was obviously wrong, as Toronto is not a US City. 

 Yet, Watson won. 

 Despite its missteps, Watson won by a huge margin -- $77,147 versus Jennings and Rutter (the two best Jeopardy players of all time) who finished with $24,000 and $21,600, respectively.   Obviously, in this case, a few instances that ‘tricked’ the model did not invalidate its ability to compete. 

So, what does this have to do with Euclidean? 

  1. Like Watson, which was fed information from books, magazines, and the web, Euclidean’s systems were fed financial statements and share prices.  
  2. Like Watson, Euclidean’s goal was to be able to evaluate new situations or questions in the context of all the available, relevant, and indexable “experience.”
  3. Both Watson and Euclidean’s systems “think” in probabilities. Watson makes statistical statements regarding whether a given answer is more or less likely to be correct.  Euclidean makes statistical statements regarding whether a company with a given operating history offered at a particular price is likely to deliver better than average returns over time. 
  4. Like Watson, which buzzes in when its confidence is sufficiently high, Euclidean aspires to buy shares in those companies with the highest confidence scores.

Also, just like when Watson answered, “What is Toronto?”, Euclidean’s models have made, and will continue to make, mistakes.  Success, however, does not require being right all of the time.  Rather, our success depends on how well our evolving portfolio of several dozen investments performs over a long period of time.  On that front, we believe the odds are on our side. 

Take Aways

When you think about Euclidean, “value traps”, and IBM’s Watson, we hope you will remember the following:

  1. We found that a fruitful method for using a company’s historical financials to evaluate it as a potential investment involves looking for all four of these qualities: high returns on capital, a long & consistent operating history, a conservative balance sheet, and a very low price. 
  2. Our evaluations of individual companies are probabilistic in nature and, therefore, imprecise. 
  3. Euclidean will make poor investments from time to time that will seem like obvious errors in retrospect. 
  4. Investment success is not about being right all of the time.  It is about operating with the odds on your side and rationally aspiring to look consistently good across long periods of time. 

In the meantime, you should know that, just like the IBM team is surely working to help Watson avoid making the same mistake twice, we are constantly searching for better practices that will reduce our errors and improve our results over time. 

This brings us to some of Euclidean’s 2011 plans.

Operational Update

As you know, we are in the midst of making a transition from Atlanta to Seattle and New York.  We will focus our R&D and technology management efforts in Seattle.   Likewise, we will focus our investor relations and administrative operations in New York.  This move should be viewed in the broader context of Euclidean evolving from a start-up business to a more established, client focused and high-growth firm. 

Why make this move? Our investors are primarily in New York, and secondarily in the Seattle and San Francisco Bay areas.  We like being close to all of you because we want you to be very comfortable with how we are managing your hard earned assets.  Also, as referenced in the prior section, we have many ideas to explore that may improve Euclidean’s performance. Pursuing these ideas requires talented associates who possess a unique combination of mathematical and programming skills.  We have great access to this talent in Seattle.  Likewise, in New York, we have access to unparalleled advisors and other business resources. 

Finally, our extended families are concentrated in Seattle and New York.   As we aspire to accomplish great things with Euclidean over the next 30+ years, we think it makes great sense to focus the business in geographies that are most aligned with our families’ goals. 

So, please expect to see both new address announcements and updated fund documents for your review this summer. 


Please let us know if you have any thoughts or comments on this letter.  We look forward to connecting with you in the weeks ahead.

Best Regards,

John & Mike