Are stocks really stronger after a split?

What "it went up" actually means, in Japanese and U.S. data

StockClub Financial research, Part 1: "A close look at stocks that split"

Data retrieved: October 2, 2026 UTC / Database split sample used in this study, Japanese stocks and U.S. market / Exploratory analysis

A stock you follow has announced a stock split. It makes sense that the price per 1 share gets lower. But can you expect anything about the price moves afterward?

What we want to check here is not only whether the price "went up after the split." Was it also strong compared with the Japanese stock market over the same period? Looking beyond the mean, at the middle of the observations and at differences by year, does the same impression hold? We will look at these in order, using the data in StockClub.

For the market comparison, we use the price of the NEXT FUNDS TOPIX Exchange Traded Fund (1306). It is an exchange-traded fund (ETF) that tracks TOPIX, and is not the index itself. Because of management fees, distributions, and tracking differences from the index, we do not treat this price comparison as the same thing as the TOPIX index or as investment returns including dividends. 1306 official information

The study covers splits of common stocks in the Japan database from 2010 to October 2, 2026. The starting point for price moves is not the announcement date but the closing price before the price adjustment date. We adjust for the price changes caused by splits and reverse splits, and do not add dividends. We look at 20, 60 and 120 trading days later, but the sample with available prices differs by window. For the U.S. market database sample used in this study, a later section adds only the post-split results.

The questions were formed after looking at the initial aggregation

The questions and angles are exploratory ones, put together after looking at the initial aggregation. This is not a study that tested a pre-registered hypothesis. Observing past price moves does not show that splits caused the price changes, nor does it give an answer on which stocks to buy from now on.

1. Are "the price went up" and "it was stronger than the market" the same?

For example, when the market as a whole is rising, the share prices of companies that split can also rise. The fact that a price went up alone does not tell us how it did compared with the market over the same period. So we first check the price moves of individual stocks, then look at the difference vs. the 1306 ETF price return over the same period.

The mean and the median are numbers that look at things differently

The mean is the sum of the price moves of all events divided by the number of events. It reflects large rises and falls as they are. The median is the middle value when the price moves are lined up from smallest to largest. With an even number of events, it is the average of the middle 2. Because different splits by the same company are each counted separately, strictly speaking it is not the "middle stock" but the "middle observation."

The price return 60 trading days after a split was a mean of +4.07% and a median of −1.94%. At 20 and 120 trading days as well, the mean is positive and the median is negative. From the rise in the mean, you cannot read that "many stocks also went up."

The next figure shows the individual stock's price return minus the 1306 price return over the same period. The unit is pp (percentage points). In a hypothetical example, if the individual stock is +5% and the ETF is +3%, the difference is +2pp. If the individual stock is −1% and the ETF is −4%, the difference is also +3pp. "A positive difference" and "the price went up" are different things.

The median difference vs. 1306 was negative at 20, 60 and 120 days. On the other hand, the mean at 60 and 120 days is positive. The mismatch seen when looking at individual stocks alone remained even after adding a benchmark. Next, we check the share of observations where the difference was positive.

The share of events that outperformed the 1306 price change was roughly 40–41%. This is the share of observed events, not a future probability or the win rate of a trading strategy. The 10th percentile of the 120-day difference was −36.72pp and the 90th percentile was +36.56pp, so the dispersion of results was also large.

1306 is an ETF, and it has cash distributions, management fees, and tracking differences from the index. In this study we do not adjust for the price drops caused by cash distributions. Because dividends on individual stocks are also not added, this difference matches neither the margin by which prices beat the TOPIX price index nor the difference in total return including dividends and distributions. 1306 official information

What we can say so far is that the price moves of individual stocks, the price difference vs. the market, and the share of positive differences each show a different side. Next, we check how the years included in the pooled aggregation differ.

2. Does it look the same when split by year?

Limiting to 60 trading days after the split, we compared by the year in which each event's ex-date (ex_date) falls. Even for splits in the same year, the observation start dates differ, and the 60-trading-day window can extend into the following year.

The mean difference had both positive and negative years. 2013 was +11.80pp and 2021 was −7.70pp. On the other hand, across the 17 years in this study, the median difference was negative in every year. Even if the way the mean looks differs by year, in some years it does not match the direction shown by the middle observations.

However, the samples are small, with 24 events in 2010 and 34 events in 2011, and 2026 is also a partial year. The mix of companies, the market environment at the time, and the excluded observations differ by year. From this figure, we cannot look for "hot years" or conclude that a policy change produced the difference in a given year.

3. What was happening at the companies that moved the mean?

To check why the mean and the median are far apart, we examined the top 3 events by price return 60 trading days after the split, in descending order. These are cases chosen after looking at the aggregation, and they are not representative examples of companies that split. First, we check whether the split ratio, dates, and price units are consistent.

Billing System's starting point of ¥90,500 is ¥905 in post-split units for the 1→100 split. 11,150 yen ÷ 905 yen − 1 gives +1,132.04%. For I'LL, 1,268 yen ÷ 2 = 634 yen is the starting point, and for Japan Communications, 21,150 yen ÷ 100 = 211.5 yen is the starting point. These are the large rises, confirmed after aligning the 1-share units before and after the split.

For Billing System, we confirmed the ratio, effective date, and monthly highs and lows in the securities report of the time (pp. 24 and 28), confirmed the dates in the JPX ex-date materials, and cross-checked the closing prices against the 96ut time series. For I'LL, we checked the securities report (pp. 19 and 22) and the JPX materials, and cross-checked the closing prices by converting the adjusted prices in the Traders October column and July column back to the units of the time. For Japan Communications, we also checked the securities report (pp. 37 and 43) and the 96ut time series.

The consistency of splits, dates, and highs and lows confirmed in primary sources and the daily closing prices cross-checked on auxiliary sites differ in the scope of confirmation. Whether the auxiliary sites and the upstream data of the research database are independent has not been confirmed. In addition, because we did not test earnings, supply and demand, or news during the period as causes, we cannot say that "this rise happened because of the split." For I'LL, the starting price for the 60 trading days before the split is missing, so we did not construct a pre-split return.

So how much does this large rise show up in the mean? The arithmetic contribution to the mean across all 2,350 events from the top 2 was +0.94pp to the price mean and +0.93pp to the mean difference vs. 1306. For the 24 events equivalent to the top 1%, they are +2.60pp and +2.51pp, respectively. The contribution here is the sum of the selected values divided by all 2,350 events. It is not a contribution as a cause, nor a share of invested funds.

Excluding the top 2, the price mean is +3.13%, and excluding the top 24 it is +1.48%, both positive. On the other hand, the mean difference vs. 1306 changes to −0.09pp and −1.70pp. How much the mean is driven by very large values, and whether the share price itself went up, need to be read separately.

The exclusion is not a treatment that judges the data to be wrong. We keep the large rises we were able to confirm in the main result for all events, and show how much the view changes. The way the top events were selected was decided after seeing the results, and this is not an analysis showing that the difference will continue in the future or that splits have an effect.

4. In the U.S. market too, are "it went up" and "it was strong" the same?

We checked the two questions seen for Japanese stocks on events in the U.S. market database as well. "US" is a classification of stocks listed in the U.S. market, and does not refer only to companies headquartered in the U.S. What we present this time is the price comparison after the split for events that met the inclusion criteria. We do not use pre-split returns for the unverified additional events.

For the comparison, we used the NEXT FUNDS TOPIX Exchange Traded Fund (1306) on the Japanese side and the State Street SPDR S&P 500 ETF Trust (SPY) on the U.S. side. SPY is an ETF that aims to track the S&P 500, which covers large U.S. stocks. Neither is the index itself; what we use here is the traded price of each ETF. Because the range and size of companies differ, we do not treat them as the same yardstick. 1306 official information, SPY official information

In both markets, we add neither individual-stock dividends nor ETF distributions, and we do not adjust for ETF ex-distribution drops. Because of management costs and tracking differences from the index, the gap cannot be restated as a difference from the TOPIX or S&P 500 index return, or from total return including dividends and distributions. We first look at individual stock prices, and then read the difference vs. the ETF.

In the main adopted U.S. market sample, the mean and median price return were positive at all of 20, 60 and 120 trading days. At 60 trading days, the mean was +1.36% and the median +1.09%. The Japanese window of the same length had a mean of +4.07% and a median of −1.94%, so the relationship between the mean and the middle observation differs. From this difference alone, we cannot decide which split stocks are better or where to invest.

Looking at the price difference vs. SPY, the mean and median for the main adopted U.S. market sample were negative at 20, 60 and 120 trading days. At 60 trading days, the difference was −1.23pp for the mean and −0.80pp for the median, and the share of observations that outperformed SPY was 46.9%. Even when the middle observation of stock prices is rising, it was not necessarily stronger than the benchmark over the same period. The question we posed with Japanese stocks changes how the results are read here as well.

Because the share of events with small split ratios also differs between markets, we also show a supplementary comparison limited to splits with a ratio of 1.5 or more. For the U.S. market at 60 trading days, there are 549 events, and the difference vs. SPY is −0.35pp for the mean and −0.43pp for the median. At 20 trading days, the mean difference was +0.04pp. We also read the means under different inclusion conditions, and we do not extend the direction of the main sample to all conditions.

Of the 29 candidates with large price switches, based on additional materials, 18 were excluded from the classification, 10 were held out pending confirmation of price quality, market, date and the like, and 1 was reinstated under limited price confirmation. There are 0 unclassified, but we did not conclude that the held-out cases are "not splits." Confirming these 29 also does not mean that the corporate actions of all U.S. events in the database were audited against primary sources.

In the case of RGC (Regencell Bioscience Holdings), which was reinstated, we confirmed from the company announcement filed with the SEC that the split was 1→38 and that the company name, identifier and ticker continued. The stored closing prices were $595.1 on June 13, 2025 and $60 on June 16. This is consistent with the display on an auxiliary price site. Adjusting the starting point once to the post-split per-share basis gives 60 ÷ (595.1 ÷ 38) − 1 = +283.13%. However, we have not retrieved the exchange's primary closing prices, and the upstream independence between the auxiliary price source and the research database is also unconfirmed.

This +283.13% is the price change on the day, across the split boundary. It is a different quantity from the 20, 60 and 120 trading-day returns. RGC's price returns at 20, 60 and 120 trading days, with the same June 13 close as the starting point, were −12.52%, −15.01% and +2.17%, respectively. It is important not to read the large one-day change as "strength 60 days after the split." All of these are descriptions of prices, not causes of the rise or results including dividends.

The following table separately shows the aggregation before the additional confirmation, the main result with RGC reinstated, and a sensitivity in which cases whose IPO date in current stock information falls after the event are held out. The last hold-out rule was added after review; it is neither a process that established the company's identity at the time nor a pre-registered condition.

There is also an additional diagnostic that leaves in 8 candidates whose price unit is unconfirmed, but those figures are for checking price conversion and are not used in this article's charts as economic returns of the main adopted sample. Including earlier diagnostics in which misclassifications were mixed in, we cannot summarize it as "the difference vs. the market is negative under every sensitivity." Inclusion or exclusion is not decided by whether performance is positive or negative; it is based on confirmation of the split, security, date and price, and we keep the reason for each update.

Because we use current product classification, companies whose data remain, and windows where prices are available, we have not been able to cover every company at the time under the same conditions. The composition of companies and the market ETFs also differ between Japan and the U.S. The results of this chapter are material from checking the same question on the database sample used in this study. They do not extend to conclusions for judging the ranking of countries, the causal effect of splits, or future price movements.

5. Does it look different when we align the same sample?

In ordinary aggregation, the longer the window, the less the prices line up, and the set of targets changes. If the result that the median is negative does not depend on how targets are chosen, would the direction be the same even when we align the before and after on the same events? So we compared the same 2,000 events for which both the 120 trading days before and after can be observed.

Here an unexpected counterexample appeared. The median stock price over the 120 days after, for the same sample, is +0.42%. In the full valid sample it was −0.48%. The explanation that the median is necessarily negative cannot be maintained. The median difference vs. 1306 is negative for the same sample as well, but it is a result separate from the movement of the stock price itself.

The trajectory calculates every point from the close before the price adjustment date. For that reason, the left end cannot be read as is as the "rate of change over the prior 120 days." The baseline differs from the before-and-after comparison in the table.

It can also be read that many companies had been rising before the split. However, this is an observation before the price adjustment date. Because the announcement time is missing for many events, we cannot say that "front-running purchases occurred before the announcement." Another sample for which after-close announcements could be confirmed is also small, limited to 37 events at 20 days later and 10 events at 60 days later. We keep it separate from the main result.

Even in a sensitivity analysis using only the first event for each company, the direction that the median difference vs. 1306 after the split is negative remained. However, this does not confirm a causal effect of splits. This is because it is not a comparison that separates prior growth, stock price rises, earnings, dividend changes and the market environment.

What this comparison shows is that the sign of the median stock price changes depending on how the sample is chosen. We have not fully resolved the question of why the group whose before and after could be observed differs from the other group.

How widely the observations are dispersed is also important in reading the conclusions. With the median alone, neither the observations that rose sharply nor those that fell can be seen.

In this figure, the thin line is the central 80% range of past observations and the thick line is the central 50% range. It summarizes the distribution while keeping all observations. There are observations outside the range shown in the figure, and the mean is calculated including them. This alone does not identify which companies or which periods moved the mean, or by how much.

6. What kind of companies are the recently split ones today?

What does the lineup of companies that split in recent years look like today? We would like to look at ROE and market cap, but if we link current values to past price movements, information that was not known at the time gets mixed in. Here we read it as a current profile separated from the price study.

We separated the financials from past price movements and compared the stored values in the database for the current cross-section retrieved on October 2, 2026. The targets are currently active common stocks: 631 companies that split between January 1, 2024 and October 2, 2026, and 3,370 other companies that are not in the split group over the same period. The other group may include companies that split before then.

In the recent-split group, the median ROE was 11.69% and the median market cap was about ¥78.7 billion. In the other group, they were 7.70% and about ¥16.5 billion. This is a limited description: the current makeup of companies that split recently differs. It is not a comparison matched on sector or size, and it is not evidence that splits raised ROE or that ROE was already high at the time of the split.

ROE measures profit relative to shareholders' equity, but the values here are measurements of values stored in the database. The current specification uses shareholders' equity from the same quarter in the current and prior year, and takes the average when conditions on balances and periods are met. It has not been audited against every company's source documents for profit attribution, consolidated versus standalone figures, or alternative bases. The missing rate also differs by group. Current values were not linked to the past price analysis, nor used as drivers of returns or as trading conditions.

DOE measures dividends relative to shareholders' equity. In StockClub's official calculation specification, it uses the actual dividend for the current fiscal year (FY) and the current- and prior-period BPS adjusted to the same share-count basis, and uses their average as the denominator when conditions are met. There are exceptions, such as when the prior-period value is unavailable. A company's DOE target is distinct from the actual DOE measured here. StockClub's DOE explainer

Current median DOE was 3.54% for the recent-split group and 2.37% for the other group. The number of valid companies and the missing rate also differ. Missing values are not replaced with 0, and companies whose stored value is 0 are included in the aggregation. These are descriptive statistics for the current cross-section, a different set of companies from those aggregated in the past price windows.

For the 615 valid companies in the recent-split group, values recalculated with the defined formula from each company's actual dividend for the FY and the adjusted BPS for the FY matched the stored DOE when rounded to the 10th decimal place. 606 companies used the average adjusted BPS of the current and prior FY, and 9 used the official period-end BPS fallback basis for cases where the prior-period value or period conditions are not met. This is not a result in which all 615 companies used average BPS. For the other group, stored values were aggregated and the formula has not been recalculated. Conditions other than the formula, such as target period, currency, and dividend attribution, and every company's primary disclosures have also not been fully audited.

The FY period-ends of the 615 companies recalculated this time range from August 2025 to July 2026, depending on the company. Because this includes changes of fiscal year-end, it is not a comparison in which all companies are converted to the same 12 months. It uses the actual dividend for the FY, and is distinguished from forecast dividends for the next period or dividends mechanically constructed from the latest quarter.

To use past ROE, DOE, and market cap to explain returns, a point-in-time check is needed, using the versions that were publicly available at the time. The differences in current profiles seen here were not used to explain the causes of splits or past or future returns.

This difference is material for describing the characteristics of today's set of companies. The question of whether pre-split financials explained subsequent returns requires a separate test using the history as it was published at the time.

7. After split news, what should you check next?

In this Japanese stock data, even where mean price returns rose, the price difference versus the market and the observations in the middle of the distribution gave a different impression. Annual means differ, and there were also counterexamples where, with the same sample, the median price return was positive. Picking a single number cannot settle whether splits as a whole are good or bad.

1. Since when is the number measured? Check the announcement date, ex-date, effective date, and price adjustment date.
2. What is it compared with? Separate the individual stock's price increase, the difference versus the ETF, and the result including dividends, and read the mean, median, and sample size together.
3. How do you check the company as it is now? Check the latest profit, shareholders' equity, dividend policy, and cash flow on the same period and basis.

If a stock interests you, check not only the split but also its disclosures on earnings and shareholder returns. StockClub's financial information and the DOE explainer are a starting point for that check. The current level of ROE or DOE cannot directly determine the direction of past or future stock prices.

Questions that remain are: which companies and which periods the observations that moved the mean are concentrated in, how much the choice of sample in which both before and after can be observed changes the results, and whether differences remain even after separating out earnings and dividend changes around the time of the split. This result does not settle explanations for these.

One sensitivity analysis using only each company's first event was checked, but repeated events by the same company and the concurrent market environment have not all been separated out. This study does not prove that splits caused price movements, nor the future direction of individual stocks.

Checking how dates and splits work

In a stock split, the number of shares increases, and the theoretical 1-share price falls in proportion to the ratio. For example, when 1 share becomes 2 shares, the share count is 2x and the theoretical price is half. The split itself does not increase the value of the total holding. It is a mechanism that lowers the amount needed for a trading unit. JPX "Stock Splits"

Dates are another important point. The announcement date, ex-date, effective date, and the price adjustment date for stock prices are not necessarily the same day. Price movements in this article are based on the "price adjustment date": the database's price_adjustment_date takes priority, and ex_date is used when it is missing. They are read separately from results based on the legal effective date and from reactions to the announcement. JPX "Ex-rights"

The difference between dates and investment units also appears in the official 2014 disclosure of Japan Communications (9424). On February 4, 2014, it announced a plan to split 1 share into 100 shares, with a record date of March 31 and an effective date of April 1. It explains that the date on which the trading unit on the exchange changes to 100 shares is March 27. The announcement, the change in trading, and the legal effect fall on different days. Japan Communications' official disclosure at the time

In this example, because a 100-share trading unit was adopted at the same time as the split, the company stated explicitly that the substantive investment unit would not change. A lower 1-share price does not necessarily mean a smaller amount required for the normal trading unit. What was confirmed in this disclosure is the mechanism and the announced schedule. Securities reports and supplementary price sites were also used to cross-check the price results above, and none of these sources is treated as a test of why prices rose.

For Japanese stocks, settlement was shortened from T+3 to T+2 for trades from July 16, 2019. For this reason, past ex-dates cannot be worked backward using only the current rules. The current date-calculation code referenced here separates the regime boundary, and for 2 stored events either side of the boundary we confirmed agreement with JPX information. The date history of all events has not been re-audited against primary sources. Japan Securities Dealers Association's materials on the rule change

Price movement calculations are adjusted for the effects of splits and reverse splits, so apparent price drops are not counted as losses. By contrast, the pre-split price bands seen later use the unadjusted closing price at the time. The height of a stock price is the price of 1 share, a different measure from market cap, which indicates company size.

It is also important not to treat the small sample whose announcement date is known and the main sample based on the price adjustment date as the same thing. Even if a stock rose before the split, that cannot be called a reaction before the announcement.

Methods, sources, and limitations

The main result of the additional U.S. chapter is the U.S. final post-conditional version v1 (after RGC reinstatement, all ratios, raw-quality conditions as the main, sample of current security information). For 20/60/120 trading days there are 1,077/1,024/964 events and 682/664/645 stocks. The main calculation uses stored unadjusted closing prices and the split and reverse-split factors within the interval, and the condition of agreement with DB-adjusted prices is a separate sensitivity analysis. Quality conditions such as missing data, volume, other corporate actions, and duplicates are applied, but we do not guarantee tradability over the whole period or a primary-source audit of all corporate actions.

The U.S. starting point is the trading day of the price adjustment based on the stored ex_date, using SPY's locally observed trading days, and computes from the t=−1 close to the t=h−1 close. Even though 20-, 60-, and 120-day windows of the same length as Japan's are shown, the trading days of each market are not the same sequence of calendar dates. It does not measure reactions to announcement dates or legal effective dates, nor yen-converted investment results.

The period range of the Japan–U.S. comparison table is 2010–2026. Limitations in identification remain due to current product classifications, surviving-company data, and the history of changes to securities and company names. Japan's financial profile remains a separate current cross-section and was not used for a point-in-time (PIT) financial comparison with the U.S. The 18 additional U.S. pre-split events are outside the verification scope and were not included in the main explanation in the text.

• Population: 2,668 split candidates classified as current common stocks recorded in the Japanese database, with an ex_date from 2010-01-01 to 2026-10-02. ETFs, REITs, and reverse splits are excluded from the main events. Companies that are no longer active are included if prices exist, but survivorship and selection bias remain from missing entries in the current master, price stoppage after delisting, and insufficient history. This is not nationwide coverage.
• Price window: price_adjustment_date takes priority, with ex_date when it is missing. t=0 is the first market-observed trading day on or after that date. Post h runs from the t=−1 close to t=h−1, and pre h from t=−h−1 to t=−1. h is 20, 60, or 120. Because trading on the price adjustment date itself is included, it cannot be read simply as "bought at the close on the split day and measured h days later".
• Calendar: based on 1306's market-observed days, the officially confirmed all-day TSE halt on 2020-10-01 is excluded from the market and stock series.
• Price adjustment: reconstructed from raw closing prices and the old-shares / new-shares factors of all splits and reverse splits, and cross-checked against the DB-adjusted series. No dividend reinvestment, trading costs, or taxes. 1306 is a proxy using the ETF price not adjusted for distribution ex-dates. The difference is the individual stock return minus the ETF return over the same interval (pp), not a sum of daily differences.
• Valid sample: post20/60/120 are 2,367/2,350/2,266 events and 1,597/1,592/1,559 stocks. Excluded by window are censoring, missing endpoint or internal prices, invalid prices, missing or zero volume at endpoints, overlap with other events, and doubts about the quality of interpolated prices. Missing values are not replaced with 0 returns. Some windows retain internal zero-volume days, and tradability over the whole period has not been verified.
• Aggregation: valid events are computed equal-weighted by event. The number of events is not the number of independent samples. There are repeated events by the same company, overlapping windows, and shared market shocks in the same period. Ordinary aggregation, the same-sample before-and-after comparison, and the first-event-per-company-only sensitivity analysis are distinguished.
• Statistical scope: price bands, ratios, years, and current sectors are exploratory comparisons. Beware of chance standouts from multiple comparisons and of small samples. This time, no causal estimation, no claim of statistical significance, and no test of trading-strategy performance was carried out. Large mean values were not removed post hoc and replaced as the main result.
• Financials: current ROE, DOE, and market cap are a separate cross-section retrieved 2026-10-02. The audit of definitions for all companies against source documents is incomplete. Returns by PIT financials and market cap, for which the history as of past publication dates is not yet guaranteed, are held out. For current DOE, an additional aggregation of stored values is included, and agreement with recalculation by the defined formula was confirmed only for the 615 companies of the recent-split group. Formula recalculation for the other group, conditions such as target period, currency, and dividend attribution, and a complete audit of all companies' primary disclosures have not been done. For the U.S., the adopted sample of the U.S. final post-conditional version v1 was added. The classification and hold-out of 29 questionable candidates are settled, but this is not a primary-source audit of all recorded events. Diagnostics with unconfirmed price basis are not included in the main result and are not generalized to the market as a whole or to the relative merits of countries.

Formula: divide the interval's closing price by its opening price and adjust by the product of the old-shares / new-shares factors of splits and reverse splits within the interval. Subtract 1 from that price multiple and multiply by 100 to get the price return (%). The difference vs. 1306 is the individual stock's return over the same interval minus 1306's price return (pp).

The source is the StockClub database's stored events, prices, and current profiles, extracted 2026-10-02 UTC. Because retrieval was sequential, it is not a snapshot taken at exactly the same moment. We do not guarantee the accuracy of every recorded event or nationwide coverage.

Primary sources: JPX's explanation of splits, explanation of ex-rights, materials on the all-day halt on October 1, 2020, 1306 official information, SPY official information, the securities reports and JPX materials for the 3 Japanese companies above, and RGC's SEC filings. The supplementary price source for each date and the scope of guarantee are stated in the relevant case.