Difference-In-Differences In Microstructure Research
Market structure changes — a tick-size pilot, a fee change — rarely hit every stock at once, and difference-in-differences uses that unevenness to separate the policy's effect from everything else moving markets at the same time.
Prerequisites: Decimalisation And Its Effects
Suppose regulators run a pilot raising tick sizes for some stocks but not others, and spreads on the treated stocks widen afterward. Simply comparing spreads before and after the change would also pick up anything else happening in markets at the same time — a volatility spike, a holiday, a broad liquidity shift — and wrongly credit or blame the policy for it.
Difference-in-differences isolates a policy's effect by comparing the before-versus-after change in the treated stocks against the before-versus-after change in untreated stocks over the same period, so market-wide noise that hits both groups cancels out.
The method needs two groups observed at two points in time: stocks subject to the tick-size pilot, and a control group of similar stocks left alone. Compute each group's change in average spread from before to after. The policy's estimated effect is the treated group's change minus the control group's change — whatever moved both groups equally (a market-wide vol event, say) drops out of that subtraction, leaving only what's distinctive to the treated group.
The approach relies on the parallel trends assumption: absent the policy, the two groups' spreads would have moved together. This is testable using pre-period data — if treated and control spreads were already diverging before the pilot started, the design is contaminated and the estimated effect is not trustworthy. The SEC's 2016–2018 Tick Size Pilot, which raised ticks for a randomly assigned subset of small-cap stocks, is a widely studied example built exactly around this design.
Related concepts
Practice in interviews
Further reading
- Angrist and Pischke, Mostly Harmless Econometrics (ch. 5)