The Oracle Portfolio - Analyzing FactSet's Buffett Strategy

by: InvestorsEdge

Summary

We review a recent FactSet Insight article that systematizes a Buffett-style trading strategy.

We find this large-cap system returned 13% annually and also performed well in most of the time frames tested.

The Oracle Strategy makes a worthy addition to our list of core strategies.

Over the weekend, I have been reading an article by FactSet's Matthew Van Der Weide entitled Systematizing the Oracle: A Quant Approach to Buffett's Investment Style. This article described Matthew's research into a trading system that attempted to mimic Warren Buffett's investment style.

I've created an InvestorsEdge strategy based loosely on Matthew's 6 rules and broken down the factors to see where the strategy's outperformance comes from. Our baseline strategy, including lots of background statistics and further graphs, charts and position data, can be found by clicking here. To access the other versions of the model mentioned in this article, simply click on the history button in the left menu and select the desired version to view.

My version of Matthews's strategy gives us the following simulated results when backtested over the last 17 years:
Equity curve This model shows a 13% return over 17 years, not bad considering that we are predominately investing in large-cap companies:

Market Cap Volatility was fairly tame over the period - you can see that the drawdowns were typically under 20% except during times of market stress. The maximum drawdown during 2008 was 45%, comfortably beating the S&P 500's 56% drop of the same period:

Drawdowns

Interestingly, the strategy handled the oil crisis and taper tantrum issues of 2015-16 without experiencing the increase in volatility seen in 2002, 2008 and 2011 (although the system wasn't as profitable during this period either).

How does the strategy work?

This strategy is based on buying companies that meet 6 rules. Our basic parameters are:

  • We will start with U.S. $1,000,000 in cash.
  • We buy all stocks that meet our investment rules.
  • The strategy rebalances its stocks on the first day of each quarter (Matthew rebalanced on a monthly basis).
  • Each transaction will include a flat fee of U.S. $7.
  • We will only consider common stocks or depository receipts with a market cap between U.S. $5bn and $15bn.

At each rebalance point, our strategy's universe is restricted to companies that meet the following additional rules:

  • Pretax Income for the past twelve months greater than US $500m.
  • P/E ratio over the last 3 years between 12 and 25.
  • Average Return on Equity over the last 3 years of 8% or greater.
  • Debt to Equity ratio less than 33%.
  • Piotroski F Score of 7 or above.

The Piotroski score, a slightly more obscure factor that seeks to identify quality companies with high profitability, liquidity and margins and low leverage, is a change to Matthew's strategy. The original called for a company to only be focused on up to 3 industries, which is information we don't carry within our dataset.

That's it - an easy-to-follow strategy that has performed well over the last 17 years. So, can we improve on it?

Breaking down the factors

We have analyzed the factors involved in selecting the universe of stocks to identify those that were adding little value:

  • The Pretax Income rule acted as a slight drag on returns - our model returns improved by 0.05% annually when this factor was removed.
  • Including stocks with a P/E ratio of 12 and above improved results, whereas selecting those with a P/E under 25 caused the model to underperform.
  • Our Return on Equity rule was the best indicator of future returns - the strategy's performance dropped dramatically without this check.
  • A company's Debt to Equity ratio had little effect on our strategy.
  • The Piotroski rule had little effect on our baseline strategy; however, it plays a major part in our revised strategy below.

Combining the best factors

So, if we just include our key performing rules we should get to improve our strategy's returns right? Wrong - here's a simulation only including stocks with our best factors, namely with a P/E greater than 12 and an average ROE of 8% at each rebalance point:

Best Rules Why didn't this work? We are looking for a collection of factors that when combined improve our probability of generating excess returns in the future, and quite simply just using two basic factors isn't enough. However, if we go back and include our Piotroski rule we get the following returns:

Best Result (with Piotroski) A dramatic improvement, mainly down to our new strategy holding positions for shorter lengths of time. Adding the Piotroski improves our timing of entering and exiting positions and also improves our win rate (from 48% to 57%).

Our volatility has also reduced from the original strategy:

Drawdowns

Will the strategy work in the future?

So is this model robust enough to trade with? One way of finding this out is to examine our models' rolling 3-year returns. This tells us how profitable the strategy would have been if we had run it for each of the 15 3-year periods going back to 2000:

Period Baseline CAGR Best Factors CAGR
2000-2 23% 30%
2001-3 15% 15%
2002-4 8% 12%
2003-5 16% 17%
2004-6 14% 13%
2005-7 14% 14%
2006-8 -5% -5%
2007-9 -2% -3%
2008-10 1% 1%
2009-11 14% 14%
2010-12 13% 12%
2011-13 15% 13%
2012-14 18% 16%
2013-15 11% 11%
2014-16 7% 7%

What this table shows us is that if we had started using either system at any time in the last 17 years it would have made money over a 3-year period, with the exception of a holding period over the financial crisis of 2007-8.

One thing you can learn from the above rolling returns is that, even though the original baseline strategy hasn't performed as well as the new one over a 17-year period, it actually beat our revised strategy in the more recent tests from 2010-2015.

What are the risks?

Investing in any trading plan involves taking on risk, and this strategy is no different. We believe that the model has 3 key risks to you as an investor:

  • Specific to designing a strategy using historical data, a key problem can be curve fitting where a model works with historical data but fails under future market conditions. To mitigate this risk we have used 3-year rolling tests to identify how small, disparate timescales have performed. If the majority of these perform well we can be more comfortable that we have developed a robust trading system that will continue to work in the future.
  • The strategy may underperform, as traditional value strategies have over the last 2-3 years. It is also not unknown for models to become less successful as more people use them (Joel Greenblatt's Magic Formula is a good example of this phenomenon). There are no guarantees in investing, but I don't believe that investing in companies with high ROE and Piotroski scores and low P/E ratios is a passing fad or will become hugely less profitable over the next few years.
  • Liquidity is often an issue for systematic strategies, as to increase returns you often have to move further down the food chain and include smaller companies than you may be comfortable with buying. This strategy is aimed at large-cap securities where liquidity should not be an issue.

Would we trade this strategy?

For us to execute a trading strategy in real life, we need to have good reason to trust that it will continue to be profitable. To help us decide the likelihood of this, we have four-high level tests that our model should pass:

  • Investable - The strategy should be tradeable in real life and should scale. Our backtests include trading fees; so frictional costs are already taken into account, and our model is profitable. Pass.
  • Intuitive - There should be logical risk- or behavioral-based reasons why the strategy works. Our model invests in companies with high Return on Equity and Piotroski scores with low Price / Earnings ratios. So we are utilizing well-known and understood factors to identify companies likely to outperform. Pass.
  • Persistent - The factors involved should work over long periods of time. Our tests have concentrated on the 2000-17 period. From an academic point of view, this would not be long enough to prove persistence. However, from our point of view at InvestorsEdge, we consider going through two major market shocks and a series of smaller downturns as enough to convince us the strategy works on a long-term basis. Pass.
  • Pervasive - Pervasiveness is more an ideal than a hard rule. Our model should work across countries, regions and sectors. We consider that a strategy that works across multiple time frames running in the country with the largest and most liquid set of exchanges in the world is acceptable. Pass.

Your takeaway

Our implementations of Matthew's strategy has shown 13-15% annual returns over a 17-year period, which is a pretty good result for a strategy that invests in large-cap stocks.

Which one would we invest in now? The original baseline seems to have performed better over the last 6-7 years. However, our "best rules" model has shown 20% returns year to date for 2017 (vs. 13% from the original). So I would have to say the new revised version just wins out.

I will be tracking this strategy in real time from 1st September and will provide monthly updates on its profitability. So stay tuned and watch this space.

Disclosure: I/we have no positions in any stocks mentioned, and no plans to initiate any positions within the next 72 hours.

I wrote this article myself, and it expresses my own opinions. I am not receiving compensation for it. I have no business relationship with any company whose stock is mentioned in this article.