Major Events and Global Markets: a cross-country, cross-industry event study

How do major macro shocks reprice stocks across industries and across countries? This study applies the event-study method (CAPM market model, abnormal and cumulative abnormal returns) to 11 industries under 5 major events, comparing the UK, US, and Chinese markets.

Course: Event study project, School of Economics and Management, Tsinghua University, December 2024. Advisor: Prof. Sun Jing. Authors: Zhu Hongyi, Marco Ortiz Togashi.

The design

Five events (event date used as day 0):

Event	Date
Global Financial Crisis (Lehman)	2008-09-15
Brexit Referendum	2016-06-23
Covid-19 lockdown in China	2020-01-23
Russia-Ukraine war	2022-02-24
Trump re-election	2024-11-06

Eleven industries: Agriculture & Food, Automotive, Defense & Aerospace, Energy & Utilities, Finance & Insurance, Healthcare & Pharmaceuticals, Manufacturing & Industrial Goods, Real Estate & Construction, Retail & Consumer Goods, Technology & Telecommunications, Transportation & Infrastructure.

For each industry-event cell, an industry return series is regressed on its market index over a pre-event estimation window to fit a CAPM market model. Expected returns are projected into a 30-day event window, the abnormal return is actual minus expected, and abnormal returns accumulate into a CAR path. OLS summaries (alpha, beta, R-squared, p-value) quantify each fit, and abnormal returns are compared across industries per event.

Findings

Industries diverge sharply in their sensitivity to events. Defense & aerospace, technology, and healthcare behaved as safe havens during geopolitical conflict and the public-health shock, while energy and manufacturing were pressured through supply-chain disruption and demand swings. The study reads these differences through policy-transmission, market-efficiency, and industry-vulnerability lenses, with cross-industry comparison per event (see figures/).

The market-model betas (from results/ols_summary.csv, built out of the 55 per-cell OLS fits) quantify the same structure and are remarkably stable across all five events: defensive industries sit well below 1, cyclical ones well above.

Industry	Brexit	Covid CN	Trump 24	GFC 08	RU war
Agriculture and Food	0.75	0.74	0.91	0.73	0.82
Automotive	1.09	1.10	1.19	1.13	1.31
Defense and Aerospace	1.03	1.09	1.22	0.94	0.91
Energy and Utilities	0.94	0.98	1.04	0.81	0.83
Finance and Insurance	1.30	1.33	1.21	1.37	1.39
Healthcare and Pharmaceuticals	0.76	0.72	0.96	0.72	0.83
Manufacturing and Industrial Goods	1.15	1.11	1.30	1.09	1.15
Real Estate and Construction	1.07	1.10	1.11	1.00	1.19
Retail and Consumer Goods	0.86	0.85	1.14	0.80	0.99
Technology and Telecommunications	0.84	0.82	0.92	0.84	0.89
Transportation and Infrastructure	1.43	1.47	1.07	1.24	1.31

Two readings worth calling out: finance carries the highest beta in almost every event (leverage amplifies shocks regardless of their nature), and defense & aerospace swings from cyclical under Trump's re-election (1.22) to defensive during the actual wars and crises (0.91-0.94), which is the safe-haven rotation in a single number.

Repository layout

src/                     analysis pipeline (run in this order)
  split_by_events.py                 split combined returns into per-event files
  expected_return_calculation.py     CAPM expected returns per industry-event
  ar_calculation.py                  abnormal return = industry - expected
  event_analysis_expected_return.py  expected-return event-study step
  event_analysis_ar_car.py           AR / CAR construction and plots
  event_analysis_ols.py              market-model OLS per industry-event
results/
  ols/                   55 OLS summaries (11 industries x 5 events): alpha, beta, R^2, p
  ols_summary.csv        the 55 cells combined into one table (industry, event, alpha, beta, ...)
  expected_returns/      55 expected-return series (CAPM projections)
  abnormal_returns/      55 abnormal-return series (industry, market, expected, AR)
  ar_car_results.csv     per-company CAR over each event window
  industry_averages.csv, industry_30day_averages.csv,
  industry_event_30day_returns_cleaned.csv   aggregated industry return panels
figures/                 5 cross-industry AR comparison plots (one per event)

Reproduce

pip install -r requirements.txt
# All scripts read their inputs from a configurable data root:
#   default   <repo>/data          (create it and drop the raw folders there)
#   override  EVENT_STUDY_DATA=/path/to/data python src/event_analysis_ols.py
# The data root must hold the raw folders the pipeline expects
# (OLS/, "UK Result/", Event_Analysis_Output/, Split_By_Industry/, ...).
# The scripts run in the order listed above.

Data

Stock and index prices were pulled from Yahoo Finance for the UK, US, and Chinese markets. Raw price data is not committed (Yahoo Finance data is not redistributable): only derived series and statistics are stored here, the CAPM expected returns, the abnormal-return series, the OLS summaries, the aggregated industry return panels, and the comparison figures. None of the committed CSVs contain raw OHLC quotes; they hold return series (percentage changes) and regression outputs only.

Notes and limitations

• The original scripts carried hard-coded absolute paths from the author's machine; these have been replaced by a configurable DATA_ROOT (EVENT_STUDY_DATA env var, default <repo>/data) with the analysis logic untouched.
• Event Analysis.py from the original working folder is not part of this project (it is an unrelated app-comparison script that happened to share the name) and is intentionally excluded.
• Estimation and event windows are short, so single-cell significance should be read alongside the cross-industry comparison rather than in isolation.

Report

The full written report (Chinese, with English abstract) is included at report/Event_Study_Report.pdf.

License

Code is released under the MIT License (see LICENSE). The written report remains the intellectual work of the two authors, Zhu Hongyi and Marco Ortiz Togashi.