📈 economics

Regulation and Firm Entry in the U.S. Federal Build-Out Era: Targeting Bias and the Limits of Industry-Exposure Identification

This paper argues that conflicting findings in the literature regarding U.S. federal regulation and firm entry stem from identification challenges, specifically targeting bias and a post-2000 collapse in cross-industry regulatory variation that weakens industry-exposure instruments, ultimately suggesting a negative association between regulation and entry rather than a definitive causal claim.

Original authors: Salman Raza

Published 2026-06-30

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Salman Raza

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

The Big Question: Do Rules Kill New Businesses?

Imagine the U.S. economy as a giant garden. For decades, economists have argued about whether adding more "fences" (federal regulations) stops new plants (new businesses) from growing.

Team A says: "Yes! More fences mean fewer new plants."
Team B says: "No! The data doesn't show a clear link."

This paper, written by Salman Raza, doesn't just pick a side. Instead, it acts like a detective inspecting the tools the other teams used to measure the garden. The main conclusion is: The tools were broken, and the garden changed shape over time.

1. The "Targeting" Trap (Why the First Look Was Wrong)

The paper starts by pointing out a major flaw in how people usually study this.

The Analogy: Imagine a city inspector who only visits the busiest, most crowded intersections to check for traffic violations.

If you just look at the data, you might say: "Wow! The busiest intersections have the most tickets!"
The Reality: The inspector didn't cause the traffic; they went there because the traffic was already heavy.

The Paper's Finding:
In the real world, government agencies tend to write rules for industries that are already big, growing, or important.

Naive View (OLS): When researchers just look at the numbers, they see that highly regulated industries actually have more new businesses. This is because the rules followed the growth, not the other way around.
The Fix (IV): The author uses a special statistical "lens" (an instrument) to separate the rules from the natural growth. When they do this, the result flips. Now, the data suggests that more rules are associated with fewer new businesses (about a 36% drop in entry for a 10% increase in rules).

The Lesson: The "positive" link seen in other studies was an illusion caused by regulators chasing active industries.

2. The "Slow-Moving" Problem (Why the Answer Isn't Simple)

Even after fixing the targeting issue, the paper warns that we can't pin down a single, perfect number.

The Analogy: Imagine trying to measure how much a slow-moving glacier is melting by looking at a river that also flows slowly.

If you try to measure the glacier's melt while also accounting for the river's natural slow flow, your math gets messy. The "trend" of the river swallows up the "trend" of the glacier.

The Paper's Finding:
Regulations don't change overnight; they build up slowly over years. New businesses also change slowly.

When the author tries to control for these slow trends (to be extra careful), the result changes.
In some models, the negative effect is strong (rules kill entry).
In other models (using very specific trend controls), the effect disappears and looks like zero.
The Conclusion: The data suggests rules likely deter entry, but because both rules and business growth move so slowly, it's impossible to get a single, clean "cause-and-effect" number. It's a "maybe, but probably yes" situation, not a "definitely yes."

3. The "Garden Changed" (Why Old Studies Don't Match New Ones)

This is the most critical part of the paper. It explains why studies from the 1980s disagree with studies from the 2010s.

The Analogy:

The 1970s–1990s (The "Build-Out" Era): Imagine a gardener planting specific, heavy fences around specific types of flowers (like roses or tulips). Some flowers get huge fences; others get none. This creates a clear difference you can measure.
The 2000s–Present (The "Accumulation" Era): Now, imagine the gardener stops building specific fences. Instead, they just sprinkle a little bit of "fence dust" evenly over the entire garden. Every flower gets the same tiny bit of dust.

The Paper's Finding:

Before 2000: Regulations were "targeted." Some industries got hit hard, others lightly. This created enough variation for scientists to see a clear effect.
After 2000: Regulations became "broad." Rules piled up evenly across almost all industries.
The Result: Because the "fence dust" is now spread so evenly, there is no longer enough difference between industries to measure the effect. The statistical "engine" that worked for the 1970s and 80s stalled after 2000. The data for recent years is too "flat" to tell us if rules are hurting new businesses today.

Summary of the Takeaways

Bias Exists: Early studies were fooled because regulators target growing industries. When you fix this, the evidence points to rules hurting new business entry.
No Perfect Number: Because rules and business growth move slowly together, we can't calculate a single, precise "damage number." The effect likely exists but is hard to isolate perfectly.
The Era Has Changed: The statistical methods that worked for the 20th century do not work for the 21st century. Since 2000, regulations have become too uniform across industries to measure their specific impact on new businesses using this type of data.

The Final Verdict: The paper doesn't give a final "guilty" or "not guilty" verdict for today's regulations. Instead, it maps out where we can see the effect (in the past) and where we are blind (in the present), explaining exactly why experts have been arguing for so long.

Technical Summary: Regulation and Firm Entry in the U.S. Federal Build-Out Era

Problem and Motivation
The literature regarding the impact of U.S. federal regulation on business dynamism is characterized by significant disagreement. Some studies find that higher regulation depresses firm entry (e.g., Bailey and Thomas, 2017), while others find no significant link (e.g., Goldschlag and Tabarrok, 2018). A central vulnerability underlying this debate is endogeneity: regulation is not assigned randomly. Agencies tend to write rules for industries that are large, growing, or politically salient. Consequently, naive estimates often confound the effects of regulation with the economic conditions that invited it, potentially leading to upward bias where more-regulated industries appear to have higher entry rates simply because they are dynamic. This paper reframes the question from seeking a clean causal claim to evaluating the identification limits of industry-exposure designs, specifically examining how targeting bias and the changing nature of regulatory variation affect empirical results.

Data and Methodology
The study utilizes a panel of 254 four-digit NAICS industries from 1978 to 2023.

Regulation Data: The primary measure of regulatory exposure comes from RegData U.S. (QuantGov), which converts the Code of Federal Regulations (CFR) into annual restriction counts and complexity metrics. Industry exposure is calculated as a probability-weighted sum of regulatory text applicability. The dataset extends to 2025 (RegData 6.0), though firm outcome data is available through 2023.
Outcome Data: Firm entry and dynamics are drawn from the Census Bureau's Business Dynamics Statistics (BDS), including establishment entry, exit, firm counts, and startup job creation.
Empirical Strategy: The baseline specification relates log entry to log regulation with industry and year fixed effects. To address endogeneity, the author employs an agency-exposure shift-share instrument (following Goldsmith-Pinkham, Sorkin, and Swift, 2020; Borusyak, Hull, and Jaravel, 2022). The instrument predicts regulation for industry $j$ by summing the product of the industry's baseline exposure share to agency $a$ (fixed at the sample start) and agency $a$ 's national rule-making in year $t$ .
Diagnostic Approach: The paper rigorously tests the validity of this design by examining shock concentration (Rotemberg weights), exclusion restrictions (subsample analyses), and the sensitivity of estimates to trend controls (addressing the "slow-moving treatment" problem identified by Meer and West, 2016).

Key Results

Evidence of Targeting Bias and Sign Reversal:
Ordinary Least Squares (OLS) estimates show a small, positive coefficient, suggesting regulators follow dynamic industries. However, the shift-share instrument reverses this sign. The instrumental variable (IV) estimate indicates that a 10% increase in regulatory exposure is associated with approximately a 3.6% decrease in establishment entry and a 3.1% decrease in firm counts. This sign reversal provides evidence consistent with upward bias in naive estimates due to regulatory targeting.
Decomposition of Entry Rates vs. Counts:
While entry rates (entry relative to the stock of establishments) often appear to rise in regulated sectors, the IV results show that absolute entry counts and total establishment counts fall. The rise in rates is a mechanical artifact of the industry base shrinking faster than the flow of new entrants. The negative association is consistent with regulatory build-out coinciding with industry contraction.
Sensitivity to Trend Controls:
The negative estimate is real but bounded and highly sensitive to the specification of trends.
- When excluding industries already in decline, the negative association persists.
- Under three-digit industry trends, the estimate remains negative but imprecise.
- Under two-digit sector trends and four-digit industry trends, the estimate attenuates to near zero.
  The paper argues that because regulatory exposure is a slow-moving, trended treatment, unit-specific trends can absorb the variation of interest, causing the estimate to range from roughly -0.44 to near zero depending on the specification.
Historical Collapse of Identifying Variation:
A critical finding is that the design's identifying power collapses after the year 2000.
- First-Stage Weakness: The first-stage F-statistic drops from 33.8 (t-statistic of 5.8) in the full sample to a weak instrument (t-statistic of 0.7) for the 2000–2023 period.
- Dispersion Collapse: The cross-industry dispersion of regulatory growth roughly halved from the 1970s to the 2000s. Regulation shifted from targeted, industry-specific "build-out" to broad accumulation that lands evenly across industries.
- Implication: The negative association is identified primarily from low-frequency variation in the twentieth century. The design cannot recover a clean estimate for the post-2000 period, explaining why studies spanning different eras reach different conclusions.
Instrument Concentration:
The shift-share instrument is heavily concentrated. The Environmental Protection Agency (EPA) carries approximately 80% of the identifying weight, with the top three agencies accounting for 87%. The effective number of agencies driving the instrument is roughly 1.5. While the association persists in non-manufacturing and non-EPA-heavy subsamples, the estimates in these subgroups are often imprecise due to weaker instruments.

Significance and Claims
The paper does not claim to provide a single, clean causal estimate of regulation's effect on entry. Instead, its significance lies in mapping the boundaries of what industry-exposure designs can and cannot identify:

Bias Confirmation: It demonstrates that targeting bias is a real and signed phenomenon in the literature, where naive estimates are likely upward-biased.
Historical Context: It explains the disagreement in prior literature as a function of the changing nature of regulation. The "build-out" era (pre-2000) offered sufficient cross-industry variation for identification, whereas the modern era of "broad accumulation" does not.
Methodological Caution: It highlights that for slow-moving treatments, standard trend controls can mechanically absorb the treatment effect, leading to a range of estimates rather than a point estimate.
Future Direction: The paper concludes that credible assessments of contemporary regulation require rule-level identification rather than cross-industry exposure, as the latter design ceases to be informative in the post-2000 period.

The study extends the sample to 2023 (and regulatory data to 2025) to confirm that the post-2000 collapse in identifying variation is robust and that the headline negative association remains a feature of the full historical sample, driven by earlier decades.