Our approach

How we study share repurchase effects

We treat every buyback announcement as a data point in a larger pattern. Our framework connects firm characteristics to market responses using methods you can inspect, challenge, and replicate.
Most commentary on repurchases stops at the headline. We keep going — into the cross-sectional variation, the control structures, and the assumptions that make or break a conclusion.
  • Documented data filters and sample selection criteria
  • Published model specifications with full replication code
  • Regular robustness checks and sensitivity analyses
View datasets Meet the team
Research methodology workspace with charts
Avg. return +2.4%

Illustrative abnormal return around buyback announcement windows, based on historical sample data.

Events Sample: 4,200+

What drives our methodological choices

#01

Replication over reputation

A finding that cannot be reproduced is not a finding. We publish code, data dictionaries, and transformation logs so others can run the same analysis independently. Our credibility depends on what happens when someone else opens the box.

#02

Transparency over tidiness

Real research is messy. We report null results, model failures, and cases where assumptions break down. A clean narrative that omits the friction is not science — it is marketing dressed in academic clothing.
#03

Specificity over generality

We avoid broad claims about buybacks as a universal signal. Instead, we map how effects vary by firm size, sector, market conditions, and program structure. Context is not a footnote — it is the main text.
#04

Collaboration over authority

We engage with practitioners and academics who challenge our methods. Every critique we receive sharpens the next version of our framework. Research improves faster when skepticism is treated as a feature.

The analytical pipeline

From raw announcement to documented conclusion

1

Universe selection

2

Event identification

3

Factor construction

4

Cross-sectional regression

5

Risk adjustment

How we build our findings

1

Universe selection

We begin by defining the population of firms and repurchase events under study. This includes setting exchange coverage, time periods, minimum market capitalization thresholds, and liquidity requirements. Every filter is documented so users understand what is in the sample and what has been excluded.

We begin by defining the population of firms and repurchase events under study. This includes setting exchange coverage, time periods, minimum market capitalization thresholds, and liquidity requirements. Every filter is documented so users understand what is in the sample and what has been excluded.

Our base universe covers primary listings with sufficient trading history. We exclude firms with incomplete financial data or ambiguous announcement timing.

Sample composition affects every conclusion that follows.

  • Exchange and market coverage rules
  • Minimum float and liquidity thresholds
  • Financial data completeness requirements
2

Event identification

We identify repurchase announcements using structured news feeds and regulatory filings. Each event is timestamped and cross-referenced against firm identifiers. We distinguish between authorization announcements, actual program initiations, and completion disclosures to avoid conflating different information events.

We identify repurchase announcements using structured news feeds and regulatory filings. Each event is timestamped and cross-referenced against firm identifiers. We distinguish between authorization announcements, actual program initiations, and completion disclosures to avoid conflating different information events.

Timing precision matters. A misaligned event date can shift the entire abnormal return profile. We verify dates against multiple sources.

Event type classification is manual for ambiguous cases.

  • Announcement source verification
  • Event date cross-validation
  • Program type classification rules
3

Factor construction

We build firm-level characteristic measures that capture repurchase program features, financial health indicators, and historical market behavior. These factors are winsorized and standardized within comparison groups to reduce outlier influence. Every transformation is logged and reproducible.

We build firm-level characteristic measures that capture repurchase program features, financial health indicators, and historical market behavior. These factors are winsorized and standardized within comparison groups to reduce outlier influence. Every transformation is logged and reproducible.

Factor construction involves choices about measurement windows, scaling variables, and handling of missing values. We document each decision point.

Winsorization thresholds are chosen before analysis begins.

  • Repurchase size and timing features
  • Financial health indicator selection
  • Standardization within sector-year groups
4

Cross-sectional regression

We estimate relationships between repurchase characteristics and subsequent outcomes using panel methods that account for time-varying effects. Our baseline specifications include sector and year fixed effects. We report coefficients, standard errors, and model fit statistics without selective presentation.

We estimate relationships between repurchase characteristics and subsequent outcomes using panel methods that account for time-varying effects. Our baseline specifications include sector and year fixed effects. We report coefficients, standard errors, and model fit statistics without selective presentation.

We test alternative specifications to assess whether conclusions are robust to different control structures and functional forms.

Standard errors are clustered at the firm level.

  • Baseline specification with fixed effects
  • Alternative control structures tested
  • Robustness to functional form changes
5

Risk adjustment

We evaluate abnormal returns using benchmark models that account for known sources of systematic variation. Our primary approach uses calendar-time portfolios to mitigate cross-correlation concerns. We also report raw returns and alternative benchmark results for comparison.

We evaluate abnormal returns using benchmark models that account for known sources of systematic variation. Our primary approach uses calendar-time portfolios to mitigate cross-correlation concerns. We also report raw returns and alternative benchmark results for comparison.

Risk adjustment is not a solved problem. We present multiple approaches and discuss where they agree and diverge.

Calendar-time portfolios reduce but do not eliminate event clustering bias.

  • Calendar-time portfolio construction
  • Fama-French factor model application
  • Alternative benchmark comparisons

How our approach differs from common shortcuts

Dimension Our approach Common shortcuts
Sample transparency
Replication materials
Null result reporting
Robustness checks
Sector-level analysis
Published limitations

Academic finance offers well-trodden paths for event study design. We follow those paths where they lead, but we also build bridges to places they do not reach.

Our abnormal return calculations use market-model benchmarks estimated over pre-event windows that exclude recent repurchase history. This prevents prior buyback effects from contaminating the baseline expected return. We report both equal-weighted and value-weighted portfolio results.

Calendar-time portfolios are our primary tool for handling cross-sectional dependence. By aggregating events into monthly holdings, we mitigate the correlation structure that plagues traditional event study inference. We report both monthly rebalanced and buy-and-hold variants.

For cross-sectional coefficient estimation, we employ Fama-MacBeth procedures that run monthly cross-sectional regressions and average coefficients through time. This approach allows risk premiums to vary while maintaining interpretable firm-level relationships.

Sensitivity analysis is not an afterthought in our pipeline. We systematically vary event windows, benchmark models, and sample restrictions to map the stability of each conclusion. Findings that evaporate under modest changes are reported as fragile, not hidden.

Our data architecture preserves every intermediate transformation. Users can trace any output back through filtering, winsorization, standardization, and aggregation steps. Nothing happens in a black box.

Research documentation materials

Explore the full documentation

Our methodology papers, replication code, and data dictionaries are available for researchers who want to understand the mechanics behind our findings or adapt our framework for their own work.

We believe research improves when others can inspect, challenge, and extend it. Our documentation is designed for analysts who want to see how conclusions are built, not just what those conclusions say.

View docs

What you will find

Detailed methodology papers explaining sample construction, model specifications, and sensitivity analyses for each research stream.
Replication code and data transformation logs that allow independent verification of published findings.
Data dictionaries describing variable definitions, measurement procedures, and handling of missing or ambiguous values.
Supplementary analyses including robustness checks, alternative specifications, and failed hypotheses we explored.
1 / 3

Cookie Preferences

We use cookies to understand how visitors interact with our research platform and to improve functionality. Essential cookies are necessary for the site to operate. You can manage your preferences below.

Manage cookies

All set

Your cookie preferences have been saved. You can update them at any time through your browser settings.

Privacy policy