I have a similar table. Slight difference in pretax underwriting profit for 2009 for some reason, but nothing material.

I'm guessing the discrepancy is because earnings figures are sometimes revised, in subsequent years. The differences seem to always be small, but for instance, in 2009, Berkshire reported $1559 in pre-tax underwriting earnings, but in 2010, they reported $1460. For almost all the reports, they give the most recent year plus the two previous years. When there was a discrepancy, I used the most recent report, assuming that the revised figure better reflected the true earnings. However, to save time, I only looked at half the annual reports, using the third most recent year as a check that I was looking at the right number (they also report post-tax underwriting earnings, you have to be careful), and revising the second-most recent earnings figure if it had changed. But since I didn't look at every year's report, I might have sometimes missed an earnings revision that would have made a small difference.

So in the case of the 2009, I have used the number reported in 2011 for 2009; you may have quite naturally used the number reported in 2010 (i.e., in the 2009 annual report.)

One thing to consider:
Due to a quirk of accounting, premiums earned take a big temporary leap upwards the year that any big retroactive insurance contract signed.
That accounts for $7.1bn in 2007 for Equitas, and $10.2bn in 2017 for AIG.
I deduct those from the "premiums earned" column before trying to estimate the average underwriting profit that way.

With that adjustment, the average underwriting profit since 2001 (your date range) has been 3.329%.

I guess that makes sense, but wouldn't that temporary leap upwards be accompanied by a drag on earnings over the next few years? That might explain why your revised numbers end up being very close to mine: 3.2% for me, 3.3% for you; as you say, nothing material.

In any case, 3% is more than enough precision; obviously, one bad 2001-like year (with -23% in pre-tax underwriting earnings) would be enough to shift the 22-year average down by a full point. 3% before tax, 2% after tax is good enough for our purposes, I would think, so I'm glad we come up with numbers that are so close.

Regards, DTB