After collecting all of the data from the N-PX filings that were available in the new form we did some evaluation of the data quality. I wrote an earlier post about some of the cases – but I want to expand on that because of some new findings.

One of the problems we have when we collect data is the challenge associated with evaluating the quality of the data. Our first goal is always answering the question – did we capture what was reported? This is not difficult to answer affirmatively with the new xml format of the N-PX filing. An equally important question is to try to evaluate for our client’s benefit – do we think the data is reasonably accurate? That takes some imagination and we do not always expect to be right. It is easy enough to identify data at the extremes in magnitude – tougher to establish whether or not data that is not seemingly extreme is accurate.

We have identified reported holdings and votes that don’t make sense. For some of the filers we have been able to reasonably (we think) decide to deflate all of the reported values by 1,000,000 shares. For example the N-PX filing made by SRH Total Return Fund (CIK: 102426) reports that they voted 1,028,000,000 shares on issues relating to Berkshire Hathaway.

Warren Buffett is almost my neighbor so I knew that number was suspicious. We pulled up their (SRH’s) NPORT-P filing made on 1/29/24 to check the number of BRK-A and BRK-B shares they reported holding as of 2/29/2024. According to that filing they held 1,028 shares of the Class A stock and 345,000 shares of the Class B.

We matched across another couple of holdings and based on that review we are reasonably confident that they inflated each of the numeric share counts in their filing by 1,000,000. Thus, when you access the database and if you pull those holdings you will see that we report 1,028 for shares voted (sharesVoted) and 1,028,000,000 for shares voted original (sharesVoted_ORG).

Unfortunately there are some cases where we cannot reasonably infer the value that should be reported. An example of this situation can be found by reviewing this filing made by Kodai Capital Management LP (CIK: 1999612). Here is a screenshot of the first three rows from the filing.

They are reporting holding more than 600,000,000 in each of 3M, Adobe and Amazon. This is a huge concentration of wealth. However, when we look at their 13-F holding reports prior to the listed meeting dates – there is no reasonable deflator we can use to systematically get back to a value close to what is reported in the 13-F filing.

We also looked at the VOTING AUTHORITY columns from the 13-F and they don’t really provide much help, I could guess that we could argue to deflate the values from the N-PX by 2,000,000 based on the values reported in the SOLE VOTING AUTHORITY column but that seems much harder to defend then the more clear-cut case from the SRH filing.

I am reluctant to toss the data (what do I know) and I am also reluctant to leave the data in the original form. Therefore, I made a decision to report None for sharesVoted, sharesOnLoan and voteRecord_sharesVoted. However, we will leave unchanged the as-reported values for sharesVoted_ORG, sharesOnLoan_ORG and voteRecord_sharesVoted_ORG.