When the share count changes, past numbers are rewritten

Scanning a company's past dividends on a stock information site is common. With several years lined up in one table, it seems to show at a glance how much the company has paid over time. Yet the amounts in that table are usually not what shareholders actually received in those years.

Jupick opened the dividend history of a leading Korean stock to check. A widely used overseas data provider records a 2016 year-end dividend of KRW 550 per share and a 2017 quarterly dividend of KRW 140 per share. Another field in the same data shows that the company carried out a 50-for-1 stock split in May 2018.

Putting the two together gives the answer. Every dividend before the split has been divided down and rewritten on the basis of the current share count. KRW 550 times 50 is KRW 27,500, and KRW 140 times 50 is KRW 7,000. The latter figures are what a shareholder holding one share at the time actually received.

It is hard to call this an error. Listing the amounts before and after the split as they were would make the dividend look as if it collapsed to one fiftieth in 2018. The problem is that nothing in the table marks these values as rewritten.

One case shows that retroactive adjustment does not stop a few years back. A large US technology company has paid dividends since 1987, and the data provider stores that first dividend as $0.000536 per share. That is less than one two-thousandth of a dollar.

The same data holds five split records for this company: 2-for-1 in 1987, 2-for-1 in 2000, 2-for-1 in 2005, 7-for-1 in 2014 and 4-for-1 in 2020. Multiplying all five gives 224.

Multiplying the stored $0.000536 by 224 gives $0.1201. That restores exactly the 12 cents the company paid each quarter at the time. In other words, the 1987 dividend on screen today is not a fact from 1987 but a value that has been passed through every split in the nearly 40 years since.

This calculation needed no outside data. Because the same provider keeps the dividend history and the split history side by side, its own data is enough to reverse the adjustment. Being able to reverse it also means the provider could have marked which values are adjusted, had it chosen to.

In the history of the Korean stock above, the trickiest stretch is just before and just after the split. The March 2018 dividend is listed as KRW 354, and the June 2018 dividend is also listed as KRW 354. The two values look identical.

But the split happened in between, in May 2018. The March KRW 354 is the KRW 17,700 paid that day divided by 50, while the June KRW 354 is the KRW 354 actually paid after the split. One is a value produced by calculation and the other is the amount as received, yet the table does not tell them apart.

This is the real cost of retroactive adjustment. It achieves a smooth time series, but at the price of losing the information on which values are observations and which are conversions. From the table alone, you cannot even tell that the company split its shares.

So when looking at a dividend history, it helps to make a habit of checking the split record alongside it. Where no split occurred, the amounts in the table can be read as they are; where a split falls in between, every value before it should be treated as converted.

Retroactive adjustment is commonly thought to happen only with stock splits. The corporate action rules that index providers publish say otherwise.

One major index provider's rules define the adjustment factor for splits and bonus issues as the share count before the issue divided by the share count after it. A split of one share into five gives a factor of 0.2. That means the one fiftieth worked out earlier is not an arbitrary convention but the industry-standard formula.

More important is the rights offering. The same rules also adjust past prices when new shares are issued to existing shareholders below the market price. They compute a theoretical ex-rights price as the average of the cum-rights price and the subscription price, weighted by the allotment ratio, and then use that value divided by the cum-rights price as the adjustment factor. The example in the rules is a $300 stock with new shares allotted at a ratio of 1 for 4 at $260, which gives a factor of 0.9733.

Here the share count did not double, and there was no split. Even so, past prices are rewritten 2.67 percent lower. The price drop caused by issuing new shares at a discount is not treated as a fall in the value of the company, so that amount is taken out. A split is not the only event that rewrites history.

One more point. While this rules document spells out a formula for each of more than 16 types of corporate action, it nowhere explains why the adjustment is made. The procedure is fully public; only the meaning is missing.

On the price side, things are one step more complicated. Jupick downloaded the closing price of the same stock for the same date twice from a widely used data tool, changing only one option. The close for August 20, 2020 came out once as $114.62 and once as $118.28.

The difference is dividends. One value has the dividends paid afterward subtracted from the past price in advance; the other reflects only splits and leaves dividends untouched. The first is the right one for computing returns that assume dividends were reinvested; the second is closer to the quoted price of that day itself.

The decisive point is that neither is the number shown on screen that day. The actual close at the time was in the $400s, because a 4-for-1 split a few days later is already reflected in both.

So if two sites show different closing prices for the same stock on the same day, it is better to check whether they use different adjustment methods before deciding that one is wrong. The unadjusted value, the split-adjusted value and the value adjusted for dividends as well all circulate under the name of that stock's closing price.

Jupick uses nine metrics to compute grades: ROE, operating margin, revenue growth, debt ratio, current ratio, interest coverage, dividend yield, payout ratio and consecutive dividend years. Eight of them are ratios and one is a count of years.

Ratios are not shaken by changes in the share count. Either the numerator and the denominator are both amounts for the whole company, or, if they are per-share amounts, the same share count sits on both sides and cancels out. Even if a stock split multiplies the share count by 50, the debt ratio and the operating margin stay the same.

Consecutive dividend years is safe for a slightly different reason. Jupick reads the cash dividend per share item from the dividend report in filings on DART, Korea's electronic disclosure system, and counts back the years in which that value is greater than zero. Because it checks only whether a dividend was paid and does not compare amounts, the result is the same whether the amount reads KRW 27,500 or KRW 550.

If Jupick instead counted whether the dividend was raised every year, as the US dividend aristocrats index does, the story would be different. A history that has gone through retroactive adjustment and rounding can contain years that look flat in the table though the dividend actually rose, or the reverse, and the streak would then describe the data processing rather than the company.

One exception needs to be noted. Market cap, the numerator of PER (price to earnings ratio) and PBR, actually grows with a rights offering. Money came into the company from issuing new shares, so the company grew along with the share count. In this case the change in the ratio is not a distortion but something that really happened, and not something to be erased by adjustment.

Dividend yield and payout ratio are stable for yet another reason. For Korean stocks, Jupick does not compute these two values itself but takes the figures the company discloses in its annual report as they are. Because the company calculates them from its own dividend and its own share price, any split has already been reflected on both sides.

Rewriting past records on a new basis happens outside finance too. In 2024, Major League Baseball added the records of more than 2,300 players who played in the Negro Leagues from 1920 to 1948 to its official record book.

The result was a change at the very top of the record book. The career batting average lead passed from the player who had long held it to another player, and the leads in slugging percentage and OPS moved as well. A name that was first yesterday is not first today.

What stands out is the method. The work came after three years of research, and the announcement also stated that about 75 percent of the game records from that period had been secured. The fact of the change and the basis on which it was made came out as a set.

Retroactive adjustment of dividend histories lacks the first part of that set. Values are changed quietly, and no mark of the change remains anywhere on screen. Whether to adjust retroactively is a choice that depends on the purpose, and neither choice is wrong. Not disclosing that a choice was made is a separate matter.

In practice, three checks filter out most of these cases. First, when looking at past dividends or prices, check whether a split or share issue occurred during that period. If one did, the earlier values are converted to the current basis.

Second, if two sites show different values, check whether their adjustment methods differ before arguing over which one is wrong. Closing prices in particular circulate as unadjusted values, split-adjusted values and values adjusted for dividends as well.

Third, metrics that compare ratios hold up better against this problem than metrics that compare amounts. Lining up how much was paid per share across many years is vulnerable to retroactive adjustment; looking at how much of that year's profit went out as dividends is not.

That is the background to why Jupick counts dividend continuity by whether a dividend was paid rather than by its amount, and grades only on eight ratios and one count of years. The share count is something a company can change, and if a grade moved with what a company can change, the grade would reflect the company's decisions rather than its condition.

This content is for information only and is not investment advice. Jupick gives no trading advice on any particular stock and does not forecast the future direction of prices or dividends, and the fact that a stock split or share issue took place is not advice of any kind. Investment decisions and responsibility for them rest with the user.