FactSet Insight - Commentary and research from our desk to yours

Systematizing the Oracle: A Quant Approach to Buffett's Investment Style

Written by Matthew Van Der Weide | Aug 25, 2017

In his previous blog post Playing the Oracle: What will Berkshire Hathaway Buy Next?, FactSet’s Bryan Adams tried to anticipate Warren Buffett’s next move by screening for companies that might be on his shopping list. Certainly, it was an ambitious task to predict not just any investor’s actions, but an investor that goes by the nickname “the Oracle.” In this post, I have lower ambitions; I’m putting on my quantitative glasses and will be looking at the past rather than the future. What I am trying to find out is how the strategy described by Bryan would have performed historically when carried out in a systematic way.

The Approach

In his post, Bryan used screening to narrow down his universe to a small set of securities to define the potential shopping list. By doing so, however, we might lose potentially useful information, or breadth, in quant language; put simply, we could use the same criteria to find companies that are unattractive by looking at the bottom of the list. When I looked at a way to convert the same criteria in a more continuous signal, another investment icon came to mind: Joseph Piotroski.

Piotroski developed a scoring system to determine the strength of a firm’s financial position. Similar to Piotroski’s famous F-Score, for each of our criteria, companies that pass will score a one, those that do not, a zero. We can then simply add up these scores to come to an overall score of 0-6.

I used the same criteria as in the previous analysis; however, I put in a cap on P/E since we don’t want to overpay. Here are the criteria:

  • Size: Market Cap between $5 and $150 billion
  • Pretax Income: Income of more than $500 million
  • P/E: Minimum P/E in the last three years between 12 and 25
  • ROE: Return on Equity above 8% in the last three years
  • DE: Debt to Equity ratio less than 33%
  • Concentration: Active in fewer than three industries

For the backtest, I looked at the MSCI All Country World, and I rebalanced my universe on a monthly basis between 2007 and 2017.

The Results

As can be seen in the chart above, the companies with the maximum score of six outperform all others. More comforting, when putting on my quant glasses again, we also observe a nice monotonically decreasing pattern, where companies with a score four or above outperform and two or below underperform the benchmark.

Playing My Own Devil’s Advocate

Relying purely on historical results to infer conclusions about the future is as scientific as we can get when it comes to quantitative investing. In my view, this also means that a good quant is very critical of the results, effectively playing his/her own devil’s advocate.

We did see good performance, but we conveniently ignored transaction costs. Taking the analysis a step further, for anyone familiar with Buffett’s style, monthly rebalancing might seem at odds with his approach, where he generally buys to hold. Therefore, a strategy with very high turnover would not be appealing.

 

Universe

     

Information

% Total

Number of

Score (Flags)

Return

Sharpe Ratio

Alpha

Beta

Ratio

Turnover

Securities

Benchmark

0.35

0.061

0.00

1.00

 

1.24

2,471.81

               

Score = 6

0.81

0.189

0.55

0.68

0.47

10.20

30.69

Score = 5

0.44

0.091

0.13

0.83

0.09

10.66

169.42

Score = 4

0.49

0.095

0.16

0.89

0.39

12.37

439.17

Score = 3

0.37

0.065

0.02

0.98

0.08

14.03

658.65

Score = 2

0.27

0.046

-0.08

1.04

-0.50

16.67

607.74

Score = 1

0.20

0.031

-0.17

1.10

-0.36

10.17

566.15

 

Table 1: Portfolio Characteristics

As we can see in the table above, the strategy is actually rather “stable”; there is relatively low turnover in all of the buckets, meaning we don’t have to churn the portfolio very often and transaction costs are manageable.

 

 

Universe Return

Annualized Universe Return

Score (Flags)

1 Month

3 Month

6 Month

1 Month

3 Month

6 Month

Benchmark

0.35

1.19

2.33

4.30

4.83

4.72

 

 

 

 

 

 

 

Score = 6

0.81

2.20

4.18

10.18

9.11

8.54

Score = 5

0.44

1.46

2.70

5.46

5.96

5.48

Score = 4

0.49

1.58

3.06

6.03

6.48

6.21

Score = 3

0.37

1.16

2.19

4.51

4.73

4.42

Score = 2

0.27

1.01

2.07

3.33

4.09

4.18

Score = 1

0.20

0.81

1.76

2.41

3.27

3.56

 

Table 2: Equal weighted portfolio returns

Given Buffett’s buy to hold philosophy, I also looked at different holding periods and there we see that with quarterly or semiannual rebalancing the signal is very persistent. We give up some performance when we look at the annualized returns, 8.54% vs. 10.18% when rebalancing on a semiannual basis, but this is still significantly higher than the benchmark return of 4.72%.

The above results assume an equal weighted portfolio of stocks, which could introduce a small cap tilt that might explain some of the outperformance as well as introduce additional turnover. Therefore, I also looked at value (market cap) weighted returns:

 

Value Weighted

 

Universe Return

Universe Return

Score (Flags)

1 Month

3 Month

6 Month

1 Month

3 Month

6 Month

Summary

0.35

1.16

2.28

4.27

4.73

4.62

 

 

 

 

 

 

 

Score = 6

0.88

2.32

4.45

11.14

9.62

9.11

Score = 5

0.47

1.66

3.25

5.77

6.82

6.61

Score = 4

0.49

1.55

3.03

6.04

6.34

6.15

Score = 3

0.34

1.17

2.30

4.10

4.77

4.65

Score = 2

0.22

0.70

1.27

2.70

2.85

2.55

Score = 1

0.06

0.47

1.22

0.77

1.87

2.45


Table 3: Value weighted portfolio returns

It seems that the weighting by market cap actually brings a small improvement in performance, but more importantly, the results are in line with the original results.

In order to test the robustness of the strategy, I ran additional versions of this test:

  • To get a more continuous signal, I converted the criteria into decile scores instead of flags and added these together
  • To see whether the performance was driven by large sector tilts, I ran a sector neutral version of the Decile Score approach
  • Similarly, I ran a region neutral version

It would be too much detail to highlight of all of these tests, but in all of these variations the signal held up and we observed very similar results.

Finally, I could not resist comparing Berkshire Hathaway’s performance to that of our quantitative approach.

A good quant acknowledges her/his limitations and admittedly, this is not a 100% apples to apples comparison. Berkshire the company consists of more than its investment portfolio, we ignored transaction costs and looked at a relatively short period of the Oracle years. With that in mind, the six scoring companies do seem to outperform Berkshire from a pure performance perspective. Perhaps more impressively, Berkshire does seem to outperform all other portfolios and is in the top between the fives and sixes.

Conclusion

As Niels Bohr once wisely said, ”it’s hard to make predictions, especially when it is about the future.” Therefore, I leave it to the reader to predict whether this is an attractive investment strategy going forward. However, we can conclude that we can take what at first might seem a very fundamental approach and execute it in a quantitative manner. Quantitative investing does not need to be a black box, and could be considered merely a means to execute a strategy in an efficient manner.

Buffett is known to invest only in companies within his circle of competence, in other words, in those companies whose business model he can understand. Whether this piece is enough to bring quant into that circle and enable Berkshire to leave its investments to a computer after his retirement remains to be seen. That prediction is definitely outside my circle, or to put it differently, only an oracle could tell.