Analytics and the Intelligent Investors: Buffett and Marks
What can we glean from their writing using text mining?
Introduction
Warren Buffett and Howard Marks, known as the Oracle of Omaha and the philosopher-king of finance, respectively, are two of the most successful investors in recent history. Buffett is the chairman of Berkshire Hathaway, a conglomerate operating in sectors including insurance, manufacturing and railroads with substantial stakes in other businesses such as Apple. Marks is the chairman of Oaktree Capital, the largest distressed securities (securities of entities approaching or in bankruptcy) investor in the world. In the investing world there are many successful people who prefer to keep their secrets to themselves. There are also many who actively self-promote in search of a following in the age of social media. Buffett and Marks lie somewhere in the middle of this spectrum as they are lucid writers who willingly share their own approach and insights about markets and the economy. Buffett’s letters have presciently touched on subjects including the dotcom bubble and the dangers of derivatives while Marks’ memos discuss an even wider range of topics.
This piece is the tl;dr for 44 years of Berkshire Hathaway shareholder letters and 31 years of memos from Howard Marks. It will look at historical trends, specific moments in time and features of the two publications. Though if you’re anywhere remotely interested in investing you should definitely go read them after reading this.
Context
If you’re still unconvinced about how educational looking at Buffett’s shareholder letters could be, here’s a graph for you:
Buffett has beaten the S&P 500 index for most of his investing career. Early on, Berkshire significantly outperformed the S&P. More recently as the conglomerate accumulated over $100 billion in cash it has converged with the index since it is naturally harder to outperform at that scale. Nevertheless from 1965–2000 Buffett’s compounded annual return averaged 20% while the index’s average return was 10.2%. So clearly there is something to be learned from his words.
Animal Spirits
The graph above is the result of a sentiment analysis done on each of Berkshire Hathaway’s shareholder letters from 1977–2020. Buffett seems to be a very optimistic person: virtually every letter has a positive sentiment score and the ones that don’t are not as low as the positive ones are high. This bullish optimism isn’t just a feature of Buffett’s letters, one of his favorite sayings is to “never bet against America”. It’s unclear why Buffett was particularly positive in 1978 but by some accounts it was a particularly good year in general. On the other hand the most negative sentiment score in 2001 has clear explanations in the bursting of the dotcom bubble and the 9/11 attacks, both of which weighed down the market. More on the Great Recession and Covid later.
In contrast to Buffett, Marks’ memos have persistently negative sentiment perhaps befitting someone focused on distressed securities. An interesting data point is the highest sentiment score being for his final 2016 memo, which analyzed Donald Trump’s surprise election victory and its potential impacts. Another thing to note is that the magnitude of the sentiment scores are higher for Marks. This is probably because he has more freedom in his memos than Buffett does in his shareholder letters, which are more broadly interesting than most CEO’s letters but are still quite sanitized.
Deep Dives: 2008 and 2020
One of Buffett’s iconic quotes is to be “fearful when others are greedy, and greedy when others are fearful.” The two graphs above, which classified words in the 2007 and 2008 shareholder letters by emotion using a different sentiment analysis show that he means what he says. In the graph based on the 2007 letter on the left you can see his anticipation score reaches a high for any of his recent letters just as the financial crisis begins. Then on the right in 2008 his fear score drops while the broader market tanked and the US entered the Great Recession. Berkshire acted on these sentiments and had a profitable crisis by providing liquidity with its massive cash pile to firms in need such as Goldman Sachs and General Electric.
Buffett’s sentiment during the pandemic-induced recession of 2020 was rather different from 2008. He was unexpectedly optimistic given the terrible circumstances as shown in the Animal Spirits section. Yet across all eight feelings his scores are lower than in 2008 perhaps reflecting his advanced age of 90. And once again these sentiment scores were an accurate reflection of his behavior with Buffett being criticized for not doing enough last year to take advantage of the circumstances as he did in 2008.
For comparison I also mined the emotions present in Marks’ final memo of 2020. The most notable difference is the much higher fear score, which makes sense given the uncertainty of 2020, which was only magnified for struggling companies of the kind Marks invests in. The other thing that stood out across the above four graphs was Marks’ distribution of emotions versus Buffett’s, perhaps providing a window into what it takes to invest in bread-and-butter equities versus distressed securities.
Words of Wisdom
Of course, there’s much more to these data than emotions so let’s dig in to what they actually talk about.
Buffett’s top word is “will”, which isn’t particularly exciting but shows a focus on the future and after all, finance theory says the value of any asset is the value of its discounted future cash flows. Other frequently appearing words include “insurance”, which is the engine of Berkshire Hathaway and “value”, with Buffett known as a “value investor” for buying relatively cheap businesses or more recently, good businesses at fair prices. The major theme of these words is a focus on businesses with top words including “business”, “businesses”, “company” and “operating”. This reflects Buffett’s reputation for focusing on the quality of a business’ operations and not fluctuations in its stock price. (Sidebar: the most correlated word with “business” in Buffett’s word cloud was “silly”).
“Will” is a close runner-up to “can” in Marks’ word cloud, which overall reveals a more standard financial vocabulary. Given the structure of Oaktree as an asset manager not a conglomerate like Berkshire, it has an obligation to focus on investors, risk and market fluctuations and Marks’ words reflect this.
Common Topics
Building up from words, a reader might want to know what are the frequent topics discussed in these publications. To find out, I ran the data through a 10-topic modeler with mixed results. Buffett’s letters defy such classification; one “topic” had this list of words: ratio, bonds, arbitrage, Scott, Fetzer, Fechheimer, GAAP, Mrs, change and profits.
Surprisingly, Marks’ memos fit into neater topics despite being fewer in number. For instance one topic had: votes, election, Clinton, Trump, vote, popular, states, Republican and win. Another had: rates, capital, interest, negative, bonds, debt, risk, yield, rate, credit. Those two could quite clearly be classified as politics and fixed-income. Similarly other topics included:
- Taxes
- Debt
- Sports
- Commodities
- Asset managers
- Investing behavior
- Equities
- Hedge funds
The caveat to this is that while the above topics may appear clear from the 30 memos used, it’s unclear how well they apply to unused or future memos.
An Oracle?
Coming into this project I was curious to know whether investor sentiment could be used not just for learning but for predicting market movements. The results of a regression analysis suggest otherwise.
Buffett’s letter sentiment score for a given year is a statistically significant predictor at a 1% significance level (see below) for Berkshire’s stock return in the previous year. This is unsurprising since the letter for a given year is written in the early part of the next year and reflects that stock performance.
What would be more interesting is if a letter’s sentiment score predicted performance in the upcoming year but Buffett’s sentiment score was not a statistically significant predictor for Berkshire’s or the S&P 500’s. The only exception is letter sentiment is a statistically significant predictor for US GDP growth at a 5% significance level. In both cases with statistically significant predictors the R-squared value is still extremely low so the takeaway is avoid relying on the Oracle’s prophecies. Given Howard Marks’ specialization in distressed investing it is less surprising that his sentiment has little relation to broader market performance.
Conclusion
This project has shown that these rich sources of investing knowledge also serve as rich sources of data-driven insights. Sentiment analysis illuminated historical trends but also the worldviews of two very different investors. Text analysis then dug deeper into what a reader would find in each letter and each memo. While they may not serve as predictors of returns, that highlights the fact that investing is not easy and it takes much more thought to be a Buffett or a Marks.
Data and Methodology
The data for this post were obtained from:
Berkshire Hathaway Shareholder Letters (1977–2020)
- The letters prior to 1977 are unavailable online
Oaktree Capital Memos from Howard Marks (1990–2020)
- Marks sometimes publishes more than one memo each year. For comparison’s sake I included only the last memo from each year
- I initially planned to analyze only Buffett’s annual shareholder letters but included Marks’s memos for two reasons:
- As an investor specialized in distressed investing he would provide a unique view that Buffett might not capture
- Buffett himself mentioned Marks’ memos as one of the few pieces of writing that he never skips because of how much he learns.
World Bank Data
All data manipulation, analysis and visualization was done using R.
About
Amjad Hamza is a sophomore in The Wharton School studying Behavioral Economics and Business Analytics. This post is for Prasanna Tambe’s class, OIDD 245: Analytics and the Digital Economy.
