Strategy family 08 / 12
Event-driven
Event-driven investing focuses on how a security's value changes around a specific corporate or market event. The events range from corporate decisions such as mergers, spin-offs, buybacks and restructurings to scheduled moments such as earnings releases, index changes and post-IPO milestones like lock-up expiries, and the family also covers trading on the probability that an event happens at all. Each situation has a timeline, a set of possible outcomes and a price that implies a view on both.
Covers
- Mergers & acquisitions
- Earnings
- Spin-offs
- Buybacks
- Restructuring
- Index changes
- Post-IPO events
- Event-probability trading
Why it is a demanding test
Events compress a great deal of reading into a narrow window: merger agreements, regulatory filings, guidance, index methodologies and prospectuses. This family tests whether a system can turn that reading into explicit probabilities and payoffs, reconcile them with what market prices imply, and update them correctly as news arrives. Outcomes are often binary, so a sound position can still lose, and evaluation has to look past single results to the quality of the probability estimates.
What we test
What a system needs to do well in this family. Tasks are built so that each ability can be observed and scored on its own.
Reading deal documents for risk
Extracting the terms that decide outcomes from merger agreements and filings: the form of consideration and any collar, conditions to closing, termination fees, financing and the regulatory approvals required.
Turning prices into implied probabilities
Backing out the market-implied probability of completion from the spread, the likely price if the deal fails and the expected timeline, and comparing it with an independent estimate. The same reasoning applies to any price that depends on whether an event occurs.
Decomposing earnings surprises and reactions
Separating the headline beat or miss from what drives the price reaction: guidance, margins, quality of earnings and how the market was positioned going in.
Estimating event-created buying and selling
Index changes, spin-off distributions and lock-up expiries create buying or selling on known dates: index funds must trade, some holders cannot keep a spun-off company, and pre-IPO shareholders become free to sell. We test whether a system can estimate the size and timing of these flows, and how much of them is already anticipated.
Valuing pieces of a restructured company
Valuing spin-offs and restructurings part by part: sum-of-the-parts, stub values, recovery for each class of claim, and how value is allocated through the capital structure.
Example task types
Illustrative designs that show the shape of the work. They are not items from the benchmark itself.
- Track: Analysis
Estimate the odds a pending merger closes
The system receives a merger agreement, the relevant regulatory filings and the price history of both companies up to a cutoff date. It must estimate the probability and expected timing of completion, the target's likely price if the deal breaks, and the annualized return the current spread offers if the deal closes on schedule. A good answer reconciles its probability with the market-implied one and explains any gap with specific evidence.
- Track: Simulation
Trade an earnings season with realistic frictions
In a historical replay of one earnings season across a universe of mid-cap stocks, the system decides which positions to hold through each release and how large to make them, using only information available beforehand. The replay charges spreads that widen around announcements and borrow costs on short positions, and applies overnight gaps that no stop order can avoid. A good answer sizes positions for the full range of possible reactions rather than the expected move alone.
- Track: Live
Forecast index changes before the announcement
Ahead of a scheduled rebalance of a rules-based equity index, the system assigns each candidate a probability of being added or removed, using only live market and fundamental data. Forecasts are locked before the official announcement and scored against it. A good answer applies the methodology precisely, including buffer rules and liquidity screens, and is honest about uncertainty where a committee has discretion.
- Track: Client
Explain a lock-up expiry to a shareholder
A user who recently bought shares in a newly listed company asks what an upcoming lock-up expiry means for them. The system must explain the mechanism, why added supply may or may not move the price, how much of the effect may already be anticipated, and how the user's horizon and position size bear on it. A good answer does not forecast the price, states the uncertainty plainly, and flags concentration risk if the holding is a large share of the user's savings.
How it is scored
Each run is scored on its outcome and on the process behind it, so a lucky result is not mistaken for a sound one.
- 01
Return versus a naive event basket
Performance is compared with a simple baseline that takes an equal-weighted position in every qualifying event, such as the target of every announced cash acquisition. Beating it consistently points to selection skill rather than exposure shared by every such event.
- 02
Tail loss on adverse outcomes
A broken deal or a disappointing quarter can erase many periods of small gains in a single event. We report the worst losses and expected shortfall, not only the average return.
- 03
Brier score on event outcomes
Stated probabilities of completion, approval, index inclusion or an earnings beat are compared with realized outcomes across many events. The score penalizes overconfidence, which matters most when a small chance of failure carries a large loss.
- 04
Term extraction from documents
Conditions, fees, dates, eligibility thresholds and other terms pulled from agreements, filings and index rules are checked against expert references. Errors are tagged by type so builders can see what to fix.
- 05
Timeline and catalyst accuracy
We compare expected dates for closings, votes, rulings and index effective dates with what happened, and check that return estimates allow for delay. Annualized returns on event trades are highly sensitive to timing.
Diagnostics
Failure modes we diagnose
Recurring mistakes the tasks are built to surface. Each one is reported separately, so builders can see which to fix first.
- 01
Pricing a pending event as done
Treating a pending deal, spin-off or index change as if it will happen, and ignoring the downside if it does not. In merger arbitrage, the loss on a broken deal is typically far larger than the spread.
- 02
Hindsight about terms and timing
Using information that was not public at decision time, such as the eventual closing date, a later increase in the offer, or restated financials in place of figures as first reported.
- 03
Treating an authorization as a purchase
Assuming an announced buyback will be executed promptly and in full. Actual repurchases are often slower and smaller than the authorization, and sometimes do not happen.
- 04
Equating an earnings beat with gains
Expecting a stock to rise because results beat consensus, without weighing guidance, the quality of the beat or how high expectations had already risen.
- 05
Ignoring seniority in a restructuring
Valuing equity in a restructuring without working through debt seniority, covenants and likely recoveries, which overstates what is left for shareholders.
Environment and data
Tasks draw on point-in-time filings, deal documents, earnings releases and transcripts, buyback announcements and later repurchase reports, index methodology documents and IPO prospectuses, with announcement times recorded as precisely as the source allows. Financial data are as first reported, consensus estimates are as they stood before each release, and deal terms reflect only amendments made public by the decision date. Simulations include price gaps through announcements, spreads that widen around events, borrow cost and availability for short legs, and the possibility that a deal is delayed, renegotiated or terminated.