The Olley-Pakes decomposition is a better diagnostic than concentration ratios but still has fundamental blind spots

The decomposition measures whether productive firms are gaining market share — a genuinely better question than “how concentrated is this market?” But it captures only static allocative efficiency, cannot detect the erosion of dynamic efficiency, and is meaningless for network goods where competition is structurally unviable.

Explanandum

How should regulators measure whether a market is functioning well? Concentration ratios and markups are poor proxies — is there something better?

Substance

The Olley-Pakes decomposition (1996) breaks industry-level productivity into two components: the unweighted average efficiency of all firms, and the covariance between firm productivity and market share. Positive covariance means productive firms are gaining share — the market’s reallocation engine is working. Albrecht argues this is superior to concentration and markup measures, which can’t distinguish efficient dominance from dysfunctional monopoly.

The decomposition works well as a negative diagnostic: when covariance is falling (as in post-2008 Britain), something is clearly wrong. But as a positive signal, it has significant limitations. High covariance today may be a legacy of past competition rather than evidence that current concentration is benign. It can be driven by demand heterogeneity, regulation, or market power itself — not just competition. And it tells you nothing about whether the reallocation mechanism is working through genuine quality competition, network effects, regulatory capture, or algorithmic demand manipulation.

Melitz and Polanec (2015) showed it handles firm entry and exit poorly — measurement bias can reach 10 percentage points over five years. It also lacks standard errors and formal inference procedures.

Supports

• The Colombian trade liberalisation case convincingly shows the decomposition capturing real competitive improvement
• The UK post-2008 case identifies a genuine pathology (zombie firms) that other metrics miss
• The AT&T equipment manufacturing data shows clear improvement after the breakup

Challenges

• The AT&T case measures only a narrow slice (equipment manufacturing) of a much larger picture that includes the destruction of Bell Labs
• Goodhart’s Law applies: if regulators optimise for Olley-Pakes covariance, it will be gamed
• For network goods, the metric is essentially meaningless since monopoly is structural
• Data requirements (firm-level productivity and market shares over time) are rarely available across all sectors

Open Questions

• Can the decomposition be extended to capture dynamic efficiency?
• Is there a way to distinguish “earned” high covariance (genuine competition) from “manufactured” high covariance (network effects, demand manipulation)?
• Should regulators treat it as a red-flag detector (low covariance = problem) rather than an all-clear signal (high covariance = healthy)?

Source Context

Central to the Albrecht article discussion. The critique emerged from examining how the metric performs across the AT&T, UK rail, and Bell Labs cases.