We develop a continuous-time model of IPO filing outcomes in which venture capitalists and entrepreneurs disagree over IPO versus M&A. The model's two withdrawal channels — distress and strategic M&A — generate equal withdrawal rates across VC and non-VC firms but a quality gap, with VC withdrawals twice as likely to earn positive risk-adjusted returns. We estimate via simulated method of moments with indirect inference, embedding IV coefficients as targeted moments. Counterfactuals identify the entrepreneurial private benefit as the dominant IPO-sustaining force.
U.S. public firms halved in three decades; VCs now back most filings.
Equal withdrawal rates, but VC exits are far higher quality.
Two withdrawal channels and a contest over control.
Causal IV evidence embedded inside a structural fit.
Counterfactuals isolate the forces holding the IPO open.
In 1996 there were 7,322 domestic firms listed on U.S. stock exchanges. By 2024 the count had fallen to roughly 3,671 — a market that quietly halved itself over a single generation.
The contraction is not subtle. Annual operating-company IPOs averaged just 113 per year from 2001 to 2025, against more than 300 per year before 2001. Filing volume fell 73% — from an average of 347 a year to 93. And as the pipeline narrowed, its composition shifted: venture-capital-backed firms now account for over 60% of filings, up from 36% in the early 1990s.
A substantial literature has tried to explain the disappearance. Some point to changing economies of scope that make listing less valuable for small firms; others to a persistent "U.S. listing gap" and the declining net benefits of public status. Regulatory cost — the compliance burden of Sarbanes-Oxley — is the popular culprit, yet recent evidence puts its share of the decline at roughly 7%. On the supply side, the deregulation of private capital after 1996 let firms raise late-stage money and simply stay private longer.
These accounts identify aggregate forces. But a central question stays open: through what economic mechanism does venture capital itself reshape the IPO landscape? This paper answers it by treating IPO filing outcomes as a laboratory.
When a firm files an IPO registration statement, it has revealed its intent to go public. What happens next — completion or withdrawal — exposes the forces that decide whether intent becomes reality.
Roughly 16% of filers withdraw, most often exiting instead through M&A. We study a sample of 592 IPO filings from 1996–2015, holding the terminal event constant — every firm is eventually acquired — so the comparison isolates the path taken, not the destination. Conditioning on the filing decision also disarms the usual selection worry: every firm in the sample has already revealed a taste for public listing.
Returns are measured by the Generalized Public Market Equivalent (GPME), which discounts each deal's terminal payoff with a stochastic discount factor estimated from public markets — so a measured quality gap reflects genuine value, not assumed risk. And the data deliver something strange.
VC-backed and non-VC firms withdraw at nearly identical rates — a gap of just 1.3 points. If VC backing signalled higher quality, they would withdraw less. It does not.
Yet VC-backed withdrawals are 2.2× as likely to earn a positive risk-adjusted return. Same rate of failure to list — wildly different consequences.
If venture-backed firms were simply better, the advantage would appear everywhere. It does not. Among firms that completed their IPO, the VC quality edge disappears entirely — 22.6% positive-GPME for VC-backed versus 22.3% for non-VC, a gap of 1.0×. The asymmetry lives only in the withdrawal channel. Whatever produces it must operate through the decision to withdraw — not through general firm quality.
The four facts together box in any explanation. Fact 1: most filers complete, a sizeable minority withdraws. Fact 2: the rate of withdrawal is the same across VC status. Fact 3: the quality of withdrawal is not. Fact 4: withdrawals cluster in cold markets — 23.6% withdraw in cold periods against 8.1% in hot ones.
If every withdrawal were failure, both groups would show uniformly negative returns. If VC firms were simply higher quality, they would withdraw less. The pattern instead demands a framework where both groups face the same types of shocks — preserving rate equality — but make different decisions conditional on those shocks. That is exactly what the model supplies.
A firm has already filed for an IPO. At every instant, the party in control may complete the offering, withdraw it for an M&A exit, or wait. Firm value drifts as a geometric Brownian motion; the market cycles between hot and cold; and control itself is contractual.
The model's engine is a contest of preferences. The entrepreneur draws a private benefit B from going public — the value of being CEO of a listed company, of liquidity, of prestige. The venture capitalist does not. That single wedge means the two parties rank IPO and M&A differently, and the contract that allocates control between them decides which ranking prevails.
At every instant the controlling party weighs three actions — complete the IPO, withdraw to M&A, or wait. Completion is the modal outcome; the model's real subject is the withdrawal, which arrives through two doors with very different economics. One is symmetric across VC status; the other is not. Together they reconcile equal rates with unequal quality.
The firm goes public — the modal outcome, taken by ~84% of filers in the data.
The firm pulls its filing and exits by trade sale — the outcome this paper dissects.
Keep the filing open and continue. Firm value keeps diffusing; the option stays alive.
Exogenous adverse shocks arrive as a Poisson process. When one hits, the firm is forced to withdraw — regardless of who holds control. The shock rate and payoff are identical for VC-backed and non-VC firms.
This channel pins the withdrawal rate: it sets a common floor neither group can fall below. Distress exits are low quality.
Acquisition offers arrive stochastically. The controlling party accepts only if the net offer beats the continuation value — and here the two parties part ways.
The VC contract specifies a threshold. While the firm performs well, the founder keeps control and her option to list. Once value slips below V*, control passes to the VC — who can redirect a struggling firm toward the M&A exit the founder, anchored by B, would inefficiently refuse.
When IPO valuations are depressed, the completion threshold rises and firms delay. Withdrawals climb through both channels.
Distress equalizes rates; the strategic channel — synergy premium η plus a lower VC reservation value — lifts conditional quality.
As cold markets grow more prevalent, the optimal threshold V* rises — more firms fall under VC control, and aggregate completion falls endogenously.
The VC quality advantage is sharpest for small firms, for same-industry acquirers, and for filings made in cold markets.
The descriptive puzzle could still be unobserved selection. To rule that out, we instrument the withdrawal decision with forces that move it through market conditions — not firm quality.
Two instruments do the work, both measured over the 60-day post-filing bookbuilding window. The first is the cumulative NASDAQ return: when markets rise after filing, investor demand strengthens and withdrawal becomes less likely. The second is the average CBOE VIX: when uncertainty spikes, underwriters pull offerings regardless of firm fundamentals. With their VC interactions, four instruments identify two endogenous variables — Withdrawn and Withdrawn × VC.
The exclusion restriction is credible because the dependent variable — the normalized GPME — is already measured relative to a stochastic discount factor. A rising market lifts terminal value but simultaneously shrinks the discount factor, leaving the risk-adjusted return roughly untouched. The instruments operate on quality only through the channel of withdrawal itself. A placebo on the non-VC subsample, where the interaction is mechanically zero, finds precisely nothing.
| Dependent: normalized GPME | (1) OLS | (2) OLS + FE | (3) 2SLS |
|---|---|---|---|
| Withdrawn | −0.305** | −0.386** | −0.594 |
| (0.152) | (0.163) | (0.605) | |
| Withdrawn × VC | 0.359* | 0.386* | 1.121* |
| (0.205) | (0.227) | (0.672) | |
| VC-backed | −0.155* | −0.179* | −0.311 |
| (0.084) | (0.093) | (0.114) | |
| Observations | 592 | 592 | 592 |
| First-stage F · Withdrawn | — | — | 16.8 |
| First-stage F · Wd × VC | — | — | 19.2 |
| Overidentification p-value | — | — | 0.723 |
| Year & industry FE · controls | No | Yes | Yes |
TABLE 01 Robust standard errors in parentheses. *** p<0.01 · ** p<0.05 · * p<0.10. First-stage F-statistics comfortably clear the weak-instrument threshold of 10; the overidentification test does not reject instrument validity.
Under 2SLS with year and industry fixed effects, the interaction Withdrawn × VC is +1.121 — significant at the 10% level, roughly three times the OLS estimate. Under a differential-beta benchmark specification it strengthens to +1.461 and crosses the 5% threshold. The VC withdrawal premium is causal — and this number does not stay a robustness check. It becomes a targeted moment inside the structural estimation, forcing the model to reproduce not just the patterns but the mechanism.
“Standard estimation cannot tell a model that explains the gap by selection from one that explains it by mechanism. Matching the causal coefficient as a moment forces the distinction.”
Nine free parameters, twelve moment conditions, three degrees of overidentification. The estimator minimizes the weighted distance between simulated and empirical moments — and one of those moments is the causal IV coefficient itself.
For every candidate parameter vector, the model's HJB system is solved, 200,000 firm paths are simulated, and twelve moments are computed: withdrawal rates, the VC quality differential, withdrawal and post-IPO timing, a within-VC size gradient, the dispersion of VC-withdrawal returns — and the 2SLS interaction coefficient. The remaining 20 parameters are calibrated from market data and held fixed. Four estimates carry the argument.
TABLE 02 All nine free parameters are significant at the 5% level; eight at the 1% level. Standard errors are sandwich-form, with a cluster-bootstrapped data-moment covariance. *** p<0.01 · ** p<0.05.
A structural model is only as credible as its moments. The estimated model matches the headline causal coefficient almost exactly — +1.033 simulated against +1.121 in the data, within one cluster-robust standard error — and tracks the VC-side quality and timing moments tightly. It is honest about its misses too: it overstates the VC share among withdrawals and pushes the quality ratio above the data's.
With the model estimated, each structural feature can be switched off in isolation. The exercise reveals which force actually sustains the public-listing decision — and the answer is not the one policy usually targets.
The counterfactuals expose a two-sided story. VC value-added creates — without the α drift, more firms decay into distress and 2.7 points of completions vanish. But the synergy premium redirects — switch off η and the VC withdrawal quality advantage collapses from a 4.7× ratio to 1.3×, while completion barely moves. VCs simultaneously hold firms together and steer the best of them quietly into M&A.
And presiding over all of it is the founder's private benefit. Remove B and completion falls off a cliff — the single largest effect in the model. The implication is pointed: if the secular IPO decline reflects an erosion of B — private secondary markets supplying liquidity, late-stage venture capital supplying prestige and capital — then the decline is, at heart, a story of weakening entrepreneurial incentives to go public.
VC management sustains +2.7pp of completions that would otherwise decay into distress.
The synergy premium is the entire quality gap — remove it and 4.7× collapses to 1.3×.
The founder's private benefit holds the listing decision open — −12pp without it.
Completion does not respond to B linearly. Drag the private benefit and watch where the curve bites — the steep stretch sits right around the estimated value, so small erosions there produce outsized losses.
At the estimated B = 0.267, the curve has nearly flattened — but just below it lies the steep descent.
FIG. 08 The curve passes through the paper's reported anchors — 77.5% at B = 0, 85.1% at B = 0.20, 89.5% at the estimate, flattening near 90%. Erosion of B near the estimate produces a sharp, threshold-like decline in IPO activity.
The counterfactuals are partial-equilibrium comparative statics: one feature moves while filing selection, acquirer behavior, and market microstructure are held fixed. A full general-equilibrium accounting — firms re-sorting into filing, acquirers responding to target supply, underwriters re-pricing — would attenuate the magnitudes.
Within the sample, the conditional completion rate barely moved across the 2001 break (83.7% before, 85.0% after). The aggregate IPO contraction is overwhelmingly a filing-rate phenomenon — which this model holds fixed. The framework speaks to which filers complete, and on what terms, not to how many firms choose to file.
Proposition IV predicts the VC quality advantage should be sharpest for small firms, for same-industry acquirers, and for cold-market filings. Splitting the 94 withdrawn filers along each dimension tests all three at once.
The data confirm every prediction. The mean-GPME gap between VC-backed and non-VC withdrawals is +0.261 across the full sample. Among small firms it widens to +0.710 and among large firms it reverses — small firms spend more time below V*, where both the α drift and the lower acceptance threshold apply. The gap is concentrated in same-industry acquisitions (+0.543) and vanishes across industries — exactly where VC acquirer networks generate synergy. And it is wider for cold-market filings, when depressed IPO valuations widen the VC's acceptance wedge.
VC value-added prevents firms from deteriorating. The entrepreneurial private benefit sustains the very incentive to go public. Market conditions decide when IPO windows open. And the VC synergy premium selectively redirects the highest-value firms toward M&A — the source of the quality differential that opened this paper. Remove any one of these forces and the IPO landscape shifts substantially; together, they account for what the data show.
Interventions aimed at filing and compliance costs miss the mechanism. The decline runs through VC exit incentives and the weakening private benefits of public status — not regulatory burden.
Venture capital simultaneously creates value and channels it toward M&A. Part of the decline is efficient reallocation — but with a real cost in public-market access to high-growth firms.
Within the model, rebuilding entrepreneurial incentives for public listing is a far more powerful lever than further reducing the regulatory cost of filing.
The IPO decline is not a market failing. It is a market reorganizing — around who controls the exit, and what they want from it.
The framework abstracts from dynamic contracting, from VC heterogeneity in matching and specialization, and from the public-market entry channels that emerged after the sample window — SPAC mergers and direct listings. Whether those developments have widened or narrowed the VC synergy premium remains an open question.