• John Hawver

Berkshire Hathaway, a Data Compounding Machine

Updated: May 8, 2019

Tomorrow is the annual Berkshire Hathaway shareholders meeting in Omaha, NE. In preparation for the meeting, I decided to do a little quantitative digging into Berkshire stock. I came upon a result that surprised me so I thought I’d share.

First, an aside or preface. Last year, for the first time, I attended the annual meeting. It’s an experience. Omaha is a great city (I was born there so I may be biased…) and the show they put on there is really something. I figured there will (sadly) only be a limited number of meetings left with both Warren and Charlie in attendance, so I made the drive from Chicago to Omaha with a very good friend of mine.

What do you get out of the meeting/show that you can’t get from the Yahoo webcast or reading the annual report? In short, interactions. A plethora of interactions with investors of all stripes. I sat in the top section of CenturyLink Center, looking down onto the stage, next to a pediatrician from Denver who bought 100 shares of Berkshire in the early ’80s. He’d never sold a share and had attended the meeting every year. I could go on.

The second thing you get is the deep impact of seeing something live and hearing the wisdom of Warren and Charlie in person. Last year Warren opened the meeting by comparing an investment in the SP500 vs an equivalent amount of gold. The difference in returns in stunning. Here’s a link to last years transcript. Warren’s main point was despite numerous recessions, wars, presidents, etc, American business is the best investment you can make. That resonates, but I also took away something else.

Gold and gold mining are a lot like trading for a living. Profitable, shiny, but ultimately they don’t compound that well. Why? Well, veins of ore run dry. Similarly, trading strategies that work tend to attract acolytes and get over-competed. Look no farther than the dearth of hedge fund returns in the last decade for evidence. Money attracts people, and trading returns attract really smart hard working people. A whole generation of quantitative researchers piled into high-frequency trading and statistical arbitrage. That mass of talent combined with increasingly commoditized technology have diminished returns. This is capitalism.

Trading returns are a function of three things: Flow, Speed, and Intelligence. In an over-competed environment, only the players with at least 2 of the 3 thrive. This is what is taking place in the market, the biggest quantitative shops are thriving, as are the biggest banks; other players are simply trying to hang on to their piece of the pie.

OK, enough HFT, back to Berkshire. Berkshire is probably one of the most over-examined firms around. In short, it is an insurance firm that transfers the available float to more capital-intensive businesses like transportation, housing, and energy. It’s a wonderful compounding machine. While digging around, I decided to see how Berkshire lined up with other companies; what group of firms can really compare to Berky?

To do this I decided to apply a grouping approach using K-means. If K-means isn’t familiar to you (here’s a link) it is an unsupervised (no left-hand side of the regression) method that groups data points around centroids based upon the features selected. I decided to look at all the firms in the SP500, comparing their market caps and average traded volume, currently and 7 years ago. The plots are below.

So what do these plots tell us? First, it is easy to make some broad comparisons; average volumes have come down and market capitalizations have expanded. Here’s where it gets more interesting, as you can see in the plots, there are a couple of groups of stocks that have separated themselves from the pack. These are stocks in those two top groups in 2012 and in the present:

This is a very small, and elite, set of stocks, so it would be a bit naive to draw strong conclusions, but there seem to be some strong trends. Tech names are maintaining and gaining dominance, those that haven’t stayed at the leading edge, have fallen behind. I’d venture to say that in another seven years Exxon won’t be in the top groupings. And then there is Berkshire, sticking out like a sore thumb.

How has Berkshire maintained its edge while not being a tech firm? I think the key is that the adjective is incorrect. These are NOT tech firms that are in the lead, these are DATA firms. These firms have made leveraging data an art form. These firms hire and use the best machine learning and data experts to grow and expand their businesses.

Berkshire doesn’t seem to fit that bill. But if we change our perspective it might. Berkshire is an insurance firm, it sells insurance contracts. To do this well it employs an army of actuaries. That term, actuary, is boring. But actuarial math, stochastic calculus, is the foundation upon which the Black-Scholes model was built. And if you’re selling insurance, you are really selling options. To do that well, you need math, AND, you need data, lots of data.

1996. That’s when Berkshire bought the full balance of GEICO, at the eve of the Internet bubble. Berkshire had already been invested in many insurance companies by then and has continued to grow that empire. You can make the argument that Warren has been well aware of the power of data for quite some time. And now, in 2019, his firm stands among the world’s leading data companies. It is no coincidence that he’s investing in Apple and Amazon right now. Data is the new, old, moat.

If you take this data “lens” and look at other decisions Warren has made they seem even sharper. The purchase of BNSF was widely derided as being too expensive. But railroads provide a wealth of information and data about what happens in the economy. Similarly his purchase of Clayton homes or Iscar. Sure, these firms are capital intensive and are a good match for Berkshire’s insurance float, but they also provide key, first-hand, information about segments of the economy.

I don’t think Berkshire employs an army of machine learning experts yet (I’m available if you’re hiring Warren!), but with Warren at the helm (the OG Quant), reading voraciously each day, I doubt Berkshire needs the added expense.

Warren Buffett and Charlie Munger have built Berkshire Hathaway into one of the world’s biggest data-compounding machines ever.

Go Berky!


##### Code to Replicate #####

# Get Data

screenDF <- read.csv(<your screen file>)

# stock cluster data

fields <- c("Ticker", "Market.Cap", "Average.Volume", 'Price', "Sector")

clusterData <- na.omit(screenDF[, fields])

# cluster fit


stockCluster <- kmeans(clusterData[, c("Market.Cap", "Average.Volume")], 6, nstart = 20)

clusterData$cluster <- stockCluster$cluster

ctable <- table(stockCluster$cluster, clusterData$Sector)

# plot

stockCluster$cluster <- as.factor(stockCluster$cluster)

ggplot(clusterData, aes(x = Market.Cap,

y = Average.Volume,

color = stockCluster$cluster,

size = clusterData$Market.Cap)) + geom_point() + ggtitle("Equity Cluster - Present")