I’m still in process of following up on Jim’s analysis, and @cwags02, and @rrr12345. Am using Morningstar data from 1990-05-31 to 2022-09-30.
I keep coming back to the following observations about a qualitative difference between pre December 1998 and post December 1998 BRK behavior. FWIW, the GenRe acquisition was reflected in BRK’s accounting in last quarter 1998. I thought I’d see what other people think of the following.
Here’s a plot of Price/Book (from Morningstar) over time, logged, i.e. log(P/B))
Blow up period 1998-1999
Clearly, something went on with a big increase in Book in last quarter 1998, more on that in a moment.
Here’s a plot of BRK’s A-share price, logged
share price (logged)
Eyeballing trend lines of log(price) against time: the trend looks steeper pre-1998 and shallower after.
Above is Morningstar price data.
I pulled down price data from Yahoo, it actually went back farther, and it show similar trend break
I don’t separately have Book data, my data is just PriceToBook = Price/Book, from Morningstar.
I can get Book by forming the ratio: Price/PriceToBook
Here’s a plot of Book
Zooming in on that big blip towards end of 1998 shows a big rise in Boook:
What’s up with that?
Here’s a quote for the 1998 annual report:
"Our gain in net worth during 1998 was 25.9 billion, which increased the per-share book value of both our Class A and Class B stock by 48.3%. (snip)
Normally, a gain of 48.3% would call for handsprings – but not this year. (snip)
To explain: Our stock sells at a large premium over book value, which means that any issuing of shares we do – whether for cash or as consideration in a merger – instantly increases our per-share book-value figure, even though we’ve earned not a dime. What happens is that we get more per-share book value in such transactions than we give up. These transactions, however, do not deliver us any immediate gain in per-share intrinsic value, because in this respect what we give and what we get are roughly equal. Charlie Munger, Berkshire’s Vice Chairman and my partner, and I can’t tell you too often (though you may feel that we try), it’s the per-share gain in intrinsic value that counts rather than the per-share gain in book value.(snip)
On December 21, we completed our 22 billion dollar acquisition of General Re Corp. (snip)
"
Comment 1: Point dislocations
Buffett explains a rise in book as basically an artifact associated with the GenRe acquisition and that it’s not related to an equivalent rise in IntrinsicValue. BRK made other significant acquisitions, e.g. GEICO, so one might expect other single point disclocations in BookValue due to these artifacts as well.
Comment 2: Trend dislocation
What’s bothering me is that there also seems to be a trend differences pre-1998 and post-1998. IOW, there seems to be more going on than just single point artifacts.
One sees this in a plot of log(Price), earlier.
One sees this in a plot of Book over time (admittedly, I infer Book from my Price/Book data, so what happens in Price is reflected in my Book):
and here it is again with log(Book)
One sees this in a plot of log(Price) vs log(Book)
One see’s a difference around log(Book)=10.5 or so.
Turns out those points to the left of 10.5 or so are from dates before Dec 1998, and to the right are after 1998. To show this, here’s a plot of data after Dec 1998:
slope=0.88036
Here’s pre-1998
slope=1.18913
Comment:
If you break the data into “before Dec 1998” and “after Dec 1998” then the regressions look much better (as one might expect, if there was one trend before and a different trend after). But what I mean by ‘better’ is not just that they’re visually nicer, but if you look at plots of the residuals, i.e. the “noise” around the regression line, then the noise is much closer to nice looking Gaussians (if you regress using all the data and look at the noise, it most definitely is not Gaussian looking). Ordinary least square regression assumes Gaussian noise. Conclusions about statistical significance of coefficients e.g. slope, or how much variance is explained i.e. the R^2, will be suspect if the Gaussian assumption is violated.
Clearly, you can still run a line no matter what the ‘noise’ is, it’s just that you can’t believe the usual verbiage about statistical signficance etc that came with assumption of Gaussian noise. Also, ordinary least squares, which gives the regression lines plotted above, is quite sensitive to outliers, and Buffett is telling us to expect outlier points as artifacts of acquisitions. I’m kind of OK with outlier points, what bothering me is a difference in trend lines.
Question: Point dislocations versus trend dislocations
Buffett explains some point dislocations in Book as artifacts associated with acquisitions. BTW, such point artifacts will affect any ordinary least square regression trend line. We could try to handle them either by using robust regression techniques, and/or by using our “domain knowledge” of BRK and do some kind of hand adjustments (it’s a pity Jim isn’t posting here, but we have other BRK experts who may want to chime in on the latter).
Perhaps someone can independently confirm this other behavior, i.e. a qualitative trend difference before Dec 1998 and after Dec 1998?
FWIW, B shares were issued in 1996, i.e. sort of near this trend break time, conceivably this may have something to do with it?
Anyway, I’m fretting about trend dislocations. Point dislocations seems a simpler issue.
