What Happens After an Activist Shows Up?

Japan's large shareholding reporting regime stops requiring filings once a stake falls to 5% or below, so nothing is observable beyond that point. We cannot tell whether a fund sold out entirely or is still sitting on 4-and-something percent. We therefore count "fell below 5%" and "confirmed sold down to 0%" separately.

How long do they stay?

For the 322 pairs with an observed entry date, we treated the 195 pairs still holding as right-censored? at the end of the data window and estimated survival with the Kaplan-Meier method. One adjustment: the four pairs whose companies were delisted are censored at their last filing date, not at the end of the window.

A year after entry, eight out of ten activists are still there. More than half have left only after 985 days (about two years and eight months) — consistent with the rule of thumb that once an activist arrives, you should expect to live with them for about three years.

If you give them what they want, do they leave sooner?

We compared the 55 pairs where compliance with a demand could be established mechanically with the remaining 219 pairs — those that either did not comply or where no evidence of compliance could be found. The exit rate was 42% in both groups (two-sided p=1.00).

Time to exit was no different either. Splitting the Kaplan-Meier curves into the two groups and running a log-rank test gives two-sided p=0.71; the median time for half to exit is 1,009 days versus 1,065 days. In both groups roughly six in ten activists were still holding and roughly eight in ten had increased their stake.

What can be said with confidence is that no correlation was found between complying with demands and the activist leaving sooner. Which companies comply is, of course, not random: it depends on how hard the demand is, the company's situation, and the fund on the other side. The 55 complying pairs and the other 219 are different populations to begin with, so this comparison alone cannot support a causal reading.

Even where compliance could be confirmed, the activist did not leave sooner — if anything, around eight in ten increased their stake, and even after the company complied it took 1,009 days (about two years and nine months) for half of those activists to exit.

The stock is already moving before the large shareholding report comes out

To reach the 5% threshold that triggers a large shareholding report, a buyer working the open market in a typical listed company needs a median 60 trading days (about three months) even when buying 25% of daily volume every single day. The stock grinds higher throughout — up +3.2% against TOPIX in the last 20 trading days alone — so by the time the report is filed, the first leg of the move is already over.

For each company with an observed entry date we took the median daily volume over the 120 trading days before entry and estimated how many trading days it would take to acquire 5% of shares outstanding, under three assumptions about how much of daily volume a buyer can take without moving the market.

Notably, the number of days barely changes with company size: buying 25% of daily volume, it is 64 trading days for companies under ¥100bn in market cap and 53 for those at ¥300bn or more. Larger companies require far more money to reach the threshold, but their volume scales with them, so the time needed does not.

The stock rises steadily through those three months of accumulation: +1.9% against TOPIX from 120 trading days before the filing, +3.2% over the final 20 days, with the slope steepening as the filing approaches. The natural reading is that the buying itself is pushing the price up, but other investors piling in behind the move cannot be separated out, so this remains to be verified.

From the boardroom's point of view, the practical implication is that three months of accumulation have already happened by the time the filing lands. That makes "the stock has been outperforming TOPIX for three months with no particular news" a potential warning sign.

It rises after disclosure too, but that part is gone within a year

Cumulative TOPIX-relative returns measured from the filing date of the first large shareholding report — the day it appears on EDINET and the market learns of it, not the date the reporting obligation arose.

186 companies can be followed out to 250 trading days.

The "announcement effect" is real and shows up in the numbers. But it is temporary, fading within a year: the median a year after disclosure is −0.3%, with winners and losers split almost evenly. These figures cover the 269 companies that stayed listed, excluding the 17 that were delisted.

What happened to those 17 companies (who bought them and how they disappeared from the market) is covered in Part 3.

Demands for higher dividends or share buybacks — "give us back the cash" — appear in the purpose-of-holding field of 100 pairs (21%) / 99 companies; 73 of those pairs ask for both.

The control group is matched one-to-one on market capitalization (greedy nearest-neighbor matching, no partner more than 2x apart; 269 pairs). All 386 companies ever held by an activist are excluded from the control pool. Each control company inherits its partner's entry date as a pseudo-entry date, and exactly the same windows and calculations are applied.

The test of the difference (Mann-Whitney?, two-sided) gives p=0.7471. Without a control group one might read "73% raised dividends after entry, by +33.3%" as an activist effect, but that would mean crediting the activist with dividend increases that companies of the same size were making anyway — a misleading reading.

Note that p=0.74 is not proof that there is no difference; it means no evidence was found of an increase beyond the control group. Claiming the two are equal would require an equivalence test?, which this article does not perform.

For buybacks, we count a quarter as having a buyback when shares outstanding are essentially unchanged while the treasury-stock balance rises. This is inferred from financial-statement data, not from the cash-flow-statement line "purchase of treasury stock" or from timely disclosures of actual repurchases. On this measure, 71% of activist-held companies bought back shares after entry versus 67% in an equal-length window before entry (McNemar test?, two-sided p=0.28). The control group went from 55% before to 63% after (p=0.025).

But these are two within-group before-and-after comparisons; placing them side by side does not establish whether the groups differ. One being significant and the other not is not a comparison of the two groups. We therefore tested the between-group difference directly, three ways. A Fisher exact test on the companies that changed status (activist-held: 39 started versus 29 stopped; control: 38 versus 20) gives two-sided p=0.37; a Mann-Whitney test on each company's change scored as +1/0/−1 gives p=0.59; a sign test within the one-to-one matched pairs gives p=0.84. None is significant, and the point estimate of the increase is actually larger in the control group (+4.4 points versus +8.0 points).

We do not use a difference-in-differences regression. As section 05 shows, the two groups had already diverged in the two fiscal years before entry (p=0.039), so an estimator that assumes parallel trends in levels cannot be relied on. None of the three tests above depends on that assumption; they compare the pattern of change directly.

This measure can undercount. If a company buys back shares and cancels them within the same quarter, the treasury-stock balance returns to where it was and no change is detected, so the 71% and 67% figures may be understated. But the same measure is applied to the control group, so the undercount cannot flip the direction of the comparison.

In short, this analysis found no evidence that shareholder returns increased because an activist arrived.

What do they ask for?

A specific demand can be read from the "purpose of holding" field of the large shareholding report in 176 of 468 pairs (38%); the rest stop at boilerplate — "pure investment, and important proposals as circumstances warrant". Where a demand is stated, the most common by far concerns directors and governance (128 pairs).

The unit of count is the (fund, company) pair. The same fund reuses the same template across filings, so counting documents would simply count the fund's template. A pair raising several demands is counted under each.

The large shareholding report is a statutory document and the purpose-of-holding field a statutory disclosure, so it is very likely that the real conversations with directors and the board go considerably further than what is written here.

The classification does not fire on keywords alone. A statement such as "we will consider the state of Corporate Governance Code compliance and the board's governance framework in deciding whether to acquire additional shares" is treated as a factor in a buying decision, not a demand. Likewise, "engage in constructive dialogue with the issuer's board, directors and management" merely names directors as a counterparty. We require a topic word to co-occur with a demand verb and explicitly exclude the contexts of consideration, dialogue and voting.

Board size and composition

We matched the director lists in the annual securities report filed before entry and the one filed after. The median gap from entry to the later filing is 700 days (about two years — two rounds of director elections), and 181 companies passed at least one annual general meeting in between.

Board size rose at 60 companies, was unchanged at 112, and fell at 70. The increases and decreases roughly cancel out, and the overall level does not move. The one-year rise in median age is most plausibly just the passage of about two years.

An activist-nominated director joined the board at 6 companies. Restricting to the 181 companies that passed at least one AGM, it is 5 (2.8%).

This analysis does not measure the ratio of outside directors, and it does not count appointments and departures. The medians are two new directors and two departures, but with a median of 700 days — two election cycles — between the two reports, ordinary turnover cannot be separated from anything else. There is no control group for directors, so we cannot say that turnover happened "because of" the activist; all this section can claim is that board size does not change and how often an activist-nominated director is actually elected.

The longer the campaign, the more it targets the directors themselves

The time dimension, by contrast, shows a pattern. We compared the purpose-of-holding wording of the 146 pairs whose stakes predate the start of the data window (the long-running, contested cases) with the 322 pairs that entered during the window.

Mentions of directors and calls for their removal are clearly more common in long-running campaigns. Shareholder proposals, EGM requests and going-private demands show no difference. The counterparty of the conversation shifts from management to the board itself — a step up the ladder.

The direction of causality cannot be separated, however. "Campaigns escalate with time" and "contested campaigns are the ones that last" are indistinguishable in this comparison. Campaigns whose demands were met end sooner and so drop out of the long-running group — a selection effect.

The gap itself is unmistakable, and it was an unexpected finding. The important line, though, is the last one: a gap in the same direction is already present in the two years before entry (two-sided p=0.039). The observed difference therefore mixes a "declined after the activist arrived" component with an "activist arrived at a declining company" component, and these data alone cannot separate the two.

Part 1 found that companies with ROE below 8% had 1.84x the odds of being activist-held (adjusted for size and ownership structure, two-sided p=0.0002). That low-profitability companies get selected is already established; the performance gap here is best read as the continuation of that story.

And, once more, this article does not establish causation: the control group is matched on market capitalization, but this is not a randomized assignment.