Event Study Guide 2026: Powerful Vietnam Market Research

TL;DR: An event study measures whether a specific event, such as a dividend announcement or an index inclusion, produced abnormal returns in a stock's price. This first article in the DataCore research series explains why a reliable event study for the Vietnamese stock market starts with clean historical price data, not with models. It walks through the minus 16.7 percent stock-dividend trap, the choice between adjusted and unadjusted prices, estimation and event windows, benchmark models, and Vietnam-specific pitfalls such as daily price limits of 7 percent on HOSE, 10 percent on HNX, and 15 percent on UPCoM (per HOSE and HNX trading rules, as of July 2026).

Event study workflow: preparing Vietnamese stock market data before modeling

What is an event study and why does this series exist?
Why does everyone assume the hard part is the model?
The minus 16.7 percent trap: when a stock dividend fools the model
Which price should you use to calculate returns?
How do you choose estimation and event windows?
Market model or market-adjusted returns: which benchmark fits an event study in Vietnam?
What Vietnam-specific pitfalls distort an event study?
How does the DataCore Historical Price dataset support this work?
Where is this series headed next?
Frequently Asked Questions
Sources

What is an event study and why does this series exist?

An event study is a quantitative research method that measures how the price of a security reacts to a specific event: an earnings release, a dividend announcement, an index reclassification, or a new regulation. The method compares the return a stock actually earned around the event with the return it would normally have been expected to earn, and calls the difference the Abnormal Return (AR). The technique was pioneered by Fama, Fisher, Jensen and Roll (1969) in the International Economic Review, and the canonical survey remains MacKinlay (1997), "Event Studies in Economics and Finance", published in the Journal of Economic Literature.

This article opens a DataCore series on building a complete event study system for the Vietnamese stock market, covering the Ho Chi Minh Stock Exchange (HOSE), the Hanoi Stock Exchange (HNX), and the Unlisted Public Company Market (UPCoM). The series follows the same sequence a real research system is developed in: understanding the data, processing it, calculating returns, building the market return, and finally computing abnormal returns. Vietnam offers no shortage of events worth studying, from the FTSE Russell upgrade of Vietnam to Secondary Emerging Market status to regulatory changes such as Circular 25/2026 on banking liquidity.

Why begin a modeling series with a long article about data? Because in an event study, the model is rarely where projects fail. The data is. Everything that follows in this guide is an argument for that claim, grounded in the practical realities of Vietnamese market data.

Why does everyone assume the hard part is the model?

There is something interesting about how most people first learn the event study method. Most references open with familiar concepts: the Market Model, the Capital Asset Pricing Model (CAPM), the Fama and French three-factor model, Abnormal Return. A quick search turns up page after page of regression formulas and statistical tests. It is easy to walk away thinking the hardest part is picking the right model.

In practice, that is not quite true. Before any regression line is fitted, before any beta coefficient is estimated, before any abnormal return is calculated, there is a quieter step working behind the scenes: preparing the data. And almost everything in an event study starts with historical price data.

When people think of stock price data, they usually picture a fairly simple table: trading date, open, high, low, close, and volume. From a purely visual standpoint, that is accurate enough. But once the data is put to work in quantitative research, especially an event study, that simple picture quickly runs into its limits.

The minus 16.7 percent trap: when a stock dividend fools the model

Say a stock's closing price on day T-1 is 100,000 VND. On day T, the company distributes a 20 percent stock dividend: for every 100 shares held, investors receive 20 new shares. Under the standard adjustment rules used by Vietnamese exchanges, the reference price on day T gets adjusted down to roughly 83,333 VND (100,000 divided by 1.2).

If you calculate the return with the ordinary formula, comparing day T's closing price to day T-1's unadjusted closing price, you get an apparent drop of nearly minus 16.7 percent. But the investor's total holdings have not lost value at all; they simply hold more shares at a proportionally lower price. This is not real market movement. It is a mechanical artifact of the increase in shares outstanding.

Candlestick price chart of daily returns used in an event study

Feed that unadjusted return series straight into an event study, and day T gets flagged as a severe negative price shock, when in reality no market event happened at all. In a method whose whole point is isolating the market's genuine reaction to an event, a distortion of this size renders the conclusion meaningless. Stock dividends and bonus share issues are common corporate actions among listed Vietnamese companies, so this is not an edge case a researcher can safely ignore; it is a recurring feature of the dataset.

Which price should you use to calculate returns?

For an event study, the first requirement is a return series for each stock. And to calculate that series, a deceptively simple question shows up immediately: which price should be used? The raw closing price, the exchange reference price, or a price already adjusted for corporate actions such as cash dividends, stock dividends, and additional share issuances?

A choice that seems minor at this very first step, like the stock dividend example above, can end up shaping the entire outcome of the study. That is also why studies analyzing the very same event can produce different results. Not because the models differ, but because the input data was already different from the start.

And price is only the first of many questions that need answering before any calculation begins. Should trading volume be taken from continuous matched-order trading only, or should negotiated (put-through) trades be included too? A single large negotiated trade between two related shareholders can spike a day's reported volume without reflecting any genuine new capital flow or shift in market sentiment.

If the goal is studying foreign investor behavior around an event, which series should be used: net foreign buy and sell volume, or the change in remaining foreign ownership room? Only once you start chasing down answers to these seemingly small questions does it become clear that a historical price table was never just a simple price table.

How do you choose estimation and event windows?

Once the return series is trustworthy, the next design decision in any event study is time. The method splits the calendar around each event into two segments. The estimation window is the period used to learn what "normal" returns look like for the stock. The event window is the period over which the market's reaction to the event is measured.

MacKinlay (1997, Journal of Economic Literature) describes the common convention for daily data: an estimation window of roughly 120 to 250 trading days that ends before the event window begins, so the event itself does not contaminate the estimate of normal performance. For the event window, researchers typically report several widths, such as one day before through one day after the announcement, five days each side, or ten days each side, to capture both information leakage before the event and delayed reaction after it.

Estimation window: long enough to estimate parameters stably, short enough that the company's risk profile has not fundamentally changed. Around 120 to 250 trading days is the textbook range for daily returns (MacKinlay, 1997).
Gap: many designs leave a buffer of 5 to 10 trading days between the estimation window and the event window to guard against leakage.
Event window: start narrow, then widen. A window that is too wide dilutes the effect and picks up unrelated news; a window that is too narrow can miss anticipation and drift.
Post-event window: used when the research question concerns longer-run drift rather than the immediate reaction.

These choices interact with the data issues above. If the estimation window contains an unadjusted stock-dividend jump like the minus 16.7 percent artifact, the estimated beta and variance are polluted, and every abnormal return computed from them inherits the error. Window design and data preparation are not separate chapters; they are the same chapter.

Market model or market-adjusted returns: which benchmark fits an event study in Vietnam?

An abnormal return is always defined relative to a benchmark of expected return. In words: the abnormal return of stock i on day t equals the actual return of stock i on day t minus the return the benchmark model expected. The two workhorse benchmarks in the literature are simple.

Market-adjusted model: expected return equals the market return. So AR_it equals R_it minus R_mt, where R_mt is the return of a market index such as the VN-Index.
Market model: expected return equals alpha_i plus beta_i times R_mt, with alpha and beta estimated by ordinary least squares over the estimation window. So AR_it equals R_it minus (alpha_i plus beta_i times R_mt).

Brown and Warner (1985, Journal of Financial Economics) showed in simulations on daily data that these simple benchmarks detect abnormal performance about as well as far more elaborate models. That result is liberating for anyone building an event study in Vietnam: the marginal payoff sits in data quality, not model exotica.

Vietnam still adds two benchmark questions of its own. First, which index represents "the market": the VN-Index for HOSE stocks, the HNX Index, or a broader composite? A financial-sector event studied against a banking-heavy index can look artificially normal. Second, thin trading. Many HNX and UPCoM tickers trade infrequently, and infrequent trading biases ordinary least squares beta estimates. The classic corrections are the Scholes and Williams (1977) and Dimson (1979) estimators, which use leads and lags of the market return to repair the bias. An event study that ignores thin trading on smaller Vietnamese tickers will systematically misstate expected returns.

What Vietnam-specific pitfalls distort an event study?

Beyond corporate-action adjustments, the microstructure of the Vietnamese market creates several traps that standard textbook recipes do not mention. Anyone running an event study on HOSE, HNX, or UPCoM data should design around at least four of them.

Daily price limits truncate returns

Vietnamese exchanges cap daily moves around the reference price: 7 percent either way on HOSE, 10 percent on HNX, and 15 percent on UPCoM under current exchange trading rules (HOSE and HNX regulations, as of July 2026). Wider bands apply on a stock's first trading day or after a suspension of more than 25 sessions: up to 20 percent on HOSE, 30 percent on HNX, and 40 percent on UPCoM. When news is strong, a stock can hit the ceiling or floor for several consecutive sessions with almost no matched volume. The true reaction is then smeared across days, so a one-day event window mechanically understates the effect. Researchers should widen windows and track consecutive limit-hit days explicitly.

Settlement timing shifts tradability

Since 29 August 2022, the Vietnam Securities Depository and Clearing Corporation (VSDC) has settled equities so that shares and cash arrive before 13:00 on day T+2, letting investors trade them in that afternoon session (VSDC notice; Vietnam News, 25 August 2022). Practitioners call this the T+2.5 cycle. For an event study using volume or order-flow reactions, this matters: the earliest day existing buyers can respond by selling is not the event day itself, which shapes how volume effects should be interpreted across the event window.

Session structure and put-through trades contaminate volume

Trading on HOSE runs through an opening auction (ATO), continuous matching, and a closing auction (ATC), alongside a separate negotiated put-through channel. As noted above, a single negotiated block between related shareholders can multiply a day's reported volume without any genuine change in sentiment. Serious volume-based event metrics separate matched-order volume from put-through volume, which is exactly why the underlying dataset must record them separately.

Foreign ownership limits distort foreign-flow signals

Many Vietnamese stocks trade at or near their foreign ownership limit. When the room is full, foreign investors cannot buy on-exchange no matter how positive the event, so net foreign buying flatlines for reasons unrelated to sentiment. An event study of foreign investor behavior must therefore read net foreign flows together with the remaining foreign room, not in isolation.

How does the DataCore Historical Price dataset support this work?

In building an event study, historical price data is the first brick of the entire system. Everything that follows, the return series, the market return, ultimately the abnormal return, is built on top of it. No statistical model, however sophisticated, can compensate for unreliable input data. That is why this series does not start with financial models, and why this first article is dedicated to the foundation.

Throughout the series, the data comes from the Historical Price Dataset on the DataCore platform. This dataset does not just store the daily trading prices of stocks listed on HOSE, HNX, and UPCoM. It also includes reference prices, price adjustment factors following corporate action events, volume and value broken out separately by matched-order and negotiated trading, and foreign investor buy, sell, and ownership-room data.

These are exactly the fields this series will draw on throughout: not to plot charts, but to calculate returns correctly, handle price-adjustment events, and isolate foreign capital flows whenever a specific problem calls for it. Each pitfall in the previous section maps to a concrete field in the dataset, which is the practical test of whether a data source is actually ready for event study research.

Where is this series headed next?

Structure of the historical price dataset behind the event study series

This series is built to follow the same sequence a real event study system is actually developed in. Rather than starting with formulas, each article tackles one concrete problem at a time: understanding the data, processing it, calculating returns, building the market return, and completing the full process for computing abnormal returns and testing their statistical significance.

The hope is that by the end of the series, readers will not just know how to run an event study. They will understand why each step in the process exists, and how the data has shaped the final result all along. The next article returns to the historical price dataset itself. Rather than treating it as just a table of stock prices, it will explore each group of information inside it, unpack what each field means, and look at why a complete historical price dataset is the difference between a defensible study and a misleading one.

Frequently Asked Questions

What is an event study in finance?

An event study is a statistical method that measures the impact of a specific event on the value of a security. It compares actual returns around the event with the returns a benchmark model expected, and interprets the difference, the abnormal return, as the market's reaction to the event. The framework dates to Fama, Fisher, Jensen and Roll (1969) and is surveyed in MacKinlay (1997).

How long should the estimation window be?

For daily returns, the common convention is roughly 120 to 250 trading days, ending before the event window starts, often with a small buffer of 5 to 10 days to avoid leakage (MacKinlay, 1997, Journal of Economic Literature). Shorter windows produce noisy parameter estimates; much longer windows risk using stale information about the firm's risk profile.

Can you run an event study on HNX or UPCoM stocks?

Yes, but thin trading must be handled explicitly. Many HNX and UPCoM tickers have frequent zero-volume days, which biases ordinary least squares beta estimates. Corrections such as the Scholes and Williams (1977) or Dimson (1979) estimators, or using the market-adjusted model, make an event study on less liquid Vietnamese tickers defensible.

Do daily price limits bias event study results in Vietnam?

They can. Because HOSE caps daily moves at 7 percent, HNX at 10 percent, and UPCoM at 15 percent (exchange trading rules, as of July 2026), a strong reaction may be spread over several limit-hit sessions. Single-day windows then understate the true effect, so researchers widen the event window and flag consecutive limit days.

What data do you need to build an event study for Vietnam?

At minimum: daily prices with corporate-action adjustment factors, reference prices, matched-order and put-through volume recorded separately, a market index series, an exchange calendar, and, for foreign-flow questions, net foreign buy and sell data plus remaining foreign ownership room. The DataCore Historical Price Dataset carries all of these fields for HOSE, HNX, and UPCoM.

Event Study Guide 2026: Building Powerful Research for Vietnam's Stock Market

Table of Contents