Do Local Merger Screens Work? Evidence from a “Local Markets Lab”

Acknowledgements. This project originated during our time at Positive Competition (now part of BRG). We are grateful to Cyril Hariton for proposing the idea of developing a more rigorous and systematic approach to evaluating local merger screening methods. We also thank Laurent Eymard for detailed feedback on earlier drafts and many generous and insightful conversations that materially improved the paper. We thank Inès Graba and Maëlle Laborde for their help with the code in the early stages of the project.

Why local merger screens matter

In many retail and consumer-facing industries, competition is local. Consumers choose between nearby supermarkets, petrol stations, pharmacies or do it yourself (DIY) stores, and distance plays a material role in shaping substitution patterns. As a result, when authorities assess the competitive effects of mergers in these sectors, the focus often shifts from national or regional aggregates to local competitive conditions.

In practice, this creates challenges for the assessment. A single transaction may involve hundreds or thousands of outlets and, potentially, just as many “local markets” (to be sure, these are typically not antitrust geographic markets, but we’ll use the shorthand throughout the article). It is rarely feasible to conduct a full unilateral-effects analysis in each one. Screening tools are therefore used to prioritise: to identify which local areas are more likely to warrant closer scrutiny.

These screens typically follow the same logic. They construct a proxy for a local market around each outlet (often called a catchment area), compute “shares” inside that area and apply a simple rule of thumb to flag markets “at risk”. A prominent example is the “40–5” rule, which flags a local market if the merged entity’s post-merger share within that market exceeds 40 percent and the change in share due to the merger exceeds 5 percentage points.

This article is not an argument against screening. Screens are often a practical necessity. The problem is more basic: we know remarkably little about how well the standard screens actually work.

Do they reliably flag markets where mergers are likely to lead to meaningful unilateral price increases? Do they systematically over-flag and pull investigations into benign areas? Or do they systematically under-flag and risk missing genuinely problematic pockets? Are some screening methods better than others? How do their performances change if we change the rule or how we define these local markets?

These are empirical questions. Answering them using real merger cases is difficult. It would require reconstructing pre-merger screens; measuring merger-induced price effects market by market and across enough cases to detect robust patterns; and dealing with noisy price data and confounding shocks.

So we took a different approach.

A model economy where we can observe the truth

Instead of starting from real-world data, we built what we call a “Local Markets Lab”: a stylised model economy designed to make normally unobservable objects observable.

In the Lab, we can:

• compute local merger screens exactly as practitioners do
• simulate mergers under a wide range of demand conditions and market structures
• crucially, observe the true merger-induced price change in each local market

The Lab is not meant to replicate any specific industry. Its value lies elsewhere: it provides a controlled environment in which the logic of screening tools can be tested cleanly, at scale, and where we can analyse what is driving error and how to reduce it.

This article focuses on intuition and practical implications for merger assessment and therefore abstracts from many technical details. Readers interested in the full model, the simulation algorithm and a more granular presentation of the results, including several figures, can consult a technical companion paper.

The basic ingredients of the model

In our model economy, consumers are distributed evenly over a square grid composed of many cells. Each grid cell contains many consumers, who choose among nearby outlets (or not to buy at all) based on prices and travel distance to each outlet. Consumers dislike higher prices and longer travel distances; they also differ in their preferences, so that each cell will have some consumers (to be sure, not many of them) buying from far-away outlets.

Outlets are placed at discrete locations on the grid and belong to firms. Across simulations, we vary both the number of outlets and whom they belong to. In some scenarios, each firm operates one outlet; in others, one firm (the acquirer, the target or a rival) operates two. In the baseline simulations, there are never more than six outlets in total, but we also consider denser configurations (with 8, 10, 15 and 20 outlets) as robustness checks.

Firms compete in prices. Pre- and post-merger prices are computed as equilibrium outcomes. The only change induced by the merger is ownership: after the merger, the acquiring firm jointly sets the prices of all merging outlets to maximise firm-wide profits. Any price change post-merger is therefore attributable to the merger and the merger only.

In this way, the Lab lets us know the exact price effect of the merger in each local market (i.e. how much the outlet at the center of the market increases its price post-merger).

How we define local markets and “risk”

A key design choice concerns how local markets are defined.

Real cases often approximate catchment areas using travel-distance or travel-time because actual consumer locations and purchasing patterns are not observed. In the Lab, we can do better. Because we observe who buys from whom, we define each outlet’s local market as the smallest area around the outlet (think of a circle) accounting for a fixed share of its sales, which is 80 percent in the baseline simulations.

This gives us true catchment areas constructed from realised purchasing behaviour rather than proxies. Using them allows us to focus on the performance of screening metrics and rules without confounding the analysis with errors in market delineation (i.e. how well isochrones or isodistances proxy catchment areas).

We then classify a local market as “truly at risk” if the merger leads to a price increase above 5 percent at the outlet at its centre.

The screening methods we test

We consider five screening methods commonly used or discussed in practice.

Presence-based methods (PB and PBV)

These methods count which outlets fall inside a catchment area. Under PB, each outlet inside counts as one. Under PBV, outlets are weighted by their total sales volumes.

Overlap-based methods (OB and OBV)

These methods allow outlets outside the focal catchment to matter as long as their own catchments overlap with it. OB uses geometric overlap; OBV weights overlaps by outlet volumes.

OB2

OB2 is a variant designed to avoid mechanically inflating overlap areas by “double counting” them, as OB and OBV do. OB2 splits the overlaps equally across the corresponding outlets.

We refer to all five collectively as catchment-based screens.

How we evaluate screen performance

For each simulated local market, each method produces shares, and these shares map into a binary prediction—“at risk” or “not at risk”—based on a rule such as 40–5. Because we also observe the true price effect, we can classify local markets as:

• true positives (flagged and truly at risk)
• false positives (flagged but not truly at risk)
• true negatives (not flagged and not truly at risk)
• false negatives (not flagged but truly at risk)

We summarise performance of each method using three intuitive measures:

• Accuracy at positives (AP): how often the method flags truly risky markets
• Accuracy at negatives (AN): how often the method does not flag truly safe markets
• Average accuracy (AA): the average of AP and AN

As a benchmark to keep in mind: a coin-flip would score (on average) 50 percent on all three metrics.

What drives price effects in the Lab?

Before turning to screens, it is useful to understand what drives price effects in the model.

As in standard unilateral-effects logic, price increases are stronger when diversion between the merging outlets is higher. In the Lab, diversion is shaped mainly by three forces:

1. Consumer price sensitivity. When consumers are more price-sensitive, demand is more elastic and unilateral effects are smaller.
2. Distance and consumer distance sensitivity. When travel is costly, competition becomes more local and outlets are weaker substitutes. All else equal, closer outlets are closer competitors.
3. Market structure. Denser markets with more nearby alternatives weaken diversion between the merging parties.

These forces vary across simulations, generating environments with very different competitive conditions, even when catchment areas look similar.

Shares, price effects and an important instability

Because we observe actual purchases, we can compute true sales shares inside each catchment. This allows us to ask a natural question: if we knew the true shares, would a share-based rule like 40–5 reliably predict price effects?

The answer is subtle.

Within a fixed environment, delta-share is highly informative. Holding demand conditions and market structure constant, a higher delta-share typically reflects closer proximity between the merging outlets and stronger pre-merger competition, and therefore maps cleanly into larger price effects.

Across environments, however, this relationship breaks down. The same delta-share can correspond to very different price effects once price sensitivity, distance frictions or the density of alternatives changes. A “5-percentage-point delta-share” does not have a stable meaning across settings.

This instability matters because screening rules like 40–5 are applied uniformly, across markets and industries that may differ substantially along precisely these dimensions.

How does the 40–5 rule perform with actual shares?

When we apply the 40–5 rule to actual catchment shares, two patterns stand out.

First, the rule generates many false positives. Even with perfect information, a large share of flagged markets do not experience a price increase above 5 percent. Accuracy at negatives is only slightly above the coin-flip benchmark.

Second, the rule rarely misses true risks. Accuracy at positives is relatively high. In other words, 40–5 fails primarily by over-flagging, not under-flagging.

This already points to a calibration problem: the two prongs of the rule do not sit comfortably together across environments.

Which methods track actual shares better?

The Lab allows us to compare each method’s implied shares against the truth.

On that metric, the standard overlap-based measures, OB/OBV, perform best: they produce implied shares closest to actual shares on average. PB/PBV are materially less accurate. OB2 sits in between.

But screens are used to predict price effects, not shares per se.

Which methods perform better under 40–5?

Under the current 40–5 calibration, a clear pattern emerges.

All methods catch true risks at broadly similar rates. Where they differ sharply is on negatives. Presence-based methods (PB and PBV) are materially better at not flagging safe markets. Overlap-based methods tend to generate many small-but-positive delta-shares that cross the 5-point threshold without supporting a large price effect, as such small overlaps do not reflect material competitive interactions between the merging parties.

This does not mean that overlap-based methods are “wrong”. The issue is how shares are translated into a binary risk flag under a uniform rule.

When do methods work best under the current rule?

A natural concern is whether the disappointing average performance reflects unrealistic scenarios. Perhaps, in the “relevant” range of environments, the methods work well?

The simulations allow us to look directly at this. Screening performance under 40–5 is driven primarily by how local demand is (i.e. the distance sensitivity parameter). In other words, by the degree of differentiation on the supply side.

When distance frictions are very strong, outlets become quasi-local monopolies. Catchments barely overlap, price effects are rare and any screen looks good because there is little to detect.

At the opposite extreme, when distance hardly matters and consumers are not very price-sensitive, almost all markets are truly at risk. Again, a screen that flags often (as ours) tends to look good.

The uncomfortable result is that the environments that look most plausible for many retail cases, where distance matters but competition between nearby outlets is still meaningful, sit between these extremes. Precisely in that intermediate range is where accuracy is mixed and over-flagging is most pronounced.

In other words, the results do not hinge on averaging across irrelevant scenarios. There is no obvious “safe zone” in which the current screens become reliably strong.

Can performance be improved?

Yes, but not mechanically.

Raising the delta-share threshold improves accuracy at negatives but increases missed risks. Shrinking catchment areas reduces spurious overlaps and improves the performance of overlap-based methods.

Allowing thresholds to be optimised or shrinking catchment areas will change the ranking of methods and allow overlap-based screens to catch up with presence-based ones.

The broader lesson is not that one particular tweak is “the answer”. It is that screen performance is not an immutable property of a metric. It is a property of the “catchment definition + share construction + threshold rule” bundle.

What should practitioners take away?

Three points stand out.

First, local merger screens are not neutral filters. The Lab suggests that, under current calibrations, they tend to err on the side of over-flagging, even with perfect information.

Second, differences between methods often reflect how they interact with the rule used to translate shares into a binary decision, not fundamental differences in how well they capture competition.

Third, and most important, there is no reason to expect a single rule-of-thumb to perform consistently across environments. Demand conditions and spatial frictions matter, and screens that ignore them will inevitably struggle.

Screens remain useful—if their limitations are understood. The Local Markets Lab provides one way of building an evidence base on those limitations and on how screening design can be improved.