Does the Reserve Planning VC Funds Use Pay Off for Their LPs?

Reserve planning: VC funds commonly hold back as much for follow-on rounds as they spend on first checks. The evidence says the size matters less than who prices each follow-on. Rounds that bring in a new investor have done better than rounds only insiders fund, so reserve for the first kind and cap the second.

A practice built on insiders’ information can still disappoint the investors who pay for it. Sapphire Partners, which backs early-stage funds, finds that most of its managers reserve about $1 for every $1 of first checks, and that reserves “can, and more frequently do, bring down fund multiples” (Sapphire Partners, 2022).

Exhibit 1. Reserve planning: VC funds face six decisions, and the evidence on each

Decision	Common practice	What the evidence says
How much to hold back	About $1 per $1 of first checks	A larger reserve lowers the median fund; it raises the mean only when follow-ons find the winners
What a follow-on dollar buys	Pro rata in the next round	About 42% as much of a company as a seed dollar (our arithmetic on Carta’s 2026 medians)
Who gets it	The best third of the portfolio	On 1,218 AngelList seed deals, follow-on results fit a world that is 98% random
Who sets the price	A new lead, or insiders in a bridge	Rounds only existing investors fund fail more and return less
When to stop	When the company stops raising	Investors grow less likely to stop as rounds accrue
What the documents say	Pro rata for Major Investors	20 days to elect; an optional clause ends the right after a skipped round

Source: Sapphire Partners (2022); Carta (2026); Othman (2019); Ewens, Rhodes-Kropf and Strebulaev (2016); Guler (2007); NVCA (2025); Moonfire (2023) and Exhibit 4. CX Cash synthesis.

Reserve planning: VC funds hold back about half, betting that insiders will know more

Fred Wilson of Union Square Ventures wrote in 2017 that his firm raises its next fund “after investing about half of a fund” (Wilson, 2017). A fund-CFO newsletter calls 40% to 50% “the number nearly every LPA and fund model template still defaults to” (The Fund CFO, 2026). The industry’s model agreement has no such figure: ILPA’s defines a follow-on as an investment made “for the purpose of preserving or enhancing the Fund’s prior investment” and leaves the amount to the manager (ILPA, 2020).

The case for holding back rests on information. In Sapphire’s words, a partnered manager “can use this proprietary information to make a better-informed investment decision”. Paul Gompers found, in 794 venture-backed firms, that investors “concentrate investments in early stage and high technology companies where informational asymmetries are highest” (Gompers, 1995), and Michael Ewens, Ramana Nanda and Matthew Rhodes-Kropf tied spray-and-pray seed investing to cheap experiments (Ewens, Nanda & Rhodes-Kropf, 2018). Small first checks buy information; the reserve acts on it.

Information has a price, because the follow-on buys the same company later and dearer. On Carta’s medians for January to March 2026, a seed round closed at a $21.6M post-money valuation, and a Series A, typically two years later, at $63M after selling about 18% of the company (Carta, 2026). The Series A price was about 2.39 times the seed price, so a Series A dollar buys about 42% as much of a company as a seed dollar (our arithmetic). It is a ratio of medians and ignores option-pool top-ups.

That ratio sets the bar. Call it the step-up test: a follow-on beats a new first check only if the companies it backs are expected to return more than the step-up times what an average first check returns. At 2.39, the partners’ information has to find companies worth 2.39 times the average seed outcome, or the fund pays more for less.

But the partners’ picks add little to the price a new lead has already set

The best public test is a platform’s: Abraham Othman, AngelList’s head of data science, took 1,218 seed investments made on AngelList from 2014 to 2017 and simulated 10,000 portfolios of ten deals a year (Othman, 2019). One policy never followed on, one followed on whenever a seed note converted at a gain, and one only when the seed stake had at least doubled.

Exhibit 2. On AngelList, following on raised the average portfolio and lowered the typical one

Policy	Mean gross TVPI	Median gross TVPI	Beat “never” in
Never follow on	1.81x	1.60x	n/a
Always follow on	1.90x	1.54x	46%
Follow on after the stake doubles	1.80x	1.59x	41% (16% tied)

Source: Othman (2019), AngelList, tables 1 and 2; gross of fees and carry. Vendor research on one platform’s deals, which the paper warns may not represent the market.

Othman’s verdict is dry: the data fit “something like a mix of 98% Totally Random with 2% Momentum”.

First checks look much the same. A May 2026 working paper by Max Sina Knicker, Jean-Philippe Bouchaud and Michael Benzaquen compared venture portfolios with random ones matched on timing, geography, sector and size, and found “no evidence that portfolio construction increases the probability of high-multiple outcomes” (Knicker, Bouchaud & Benzaquen, 2026). It measures multiples from later rounds rather than exits and is not yet peer-reviewed.

Neither study says investors know nothing. On AngelList’s deals, the next round’s price seems to have carried most of what insiders knew: following every converting note moved the mean by only 0.09x. That is one platform’s seed deals from 2014 to 2017, and the paper warns against reading it as the market. The cost falls on the typical outcome, since a fund that reserves half its capital writes half as many first checks.

But where only insiders fund the round, reserves lose money

Michael Ewens, Matthew Rhodes-Kropf and Ilya Strebulaev compared rounds in which only existing investors participate with rounds that bring in a new investor. The inside rounds “lead to a higher likelihood of failure, lower probability of IPOs, and lower cash on cash multiples”, and “appear to be negative NPV, suggesting that investors make inefficient continuation decisions” (Ewens, Rhodes-Kropf & Strebulaev, 2016). Their explanation is the fund’s own clock: “an agency problem driven by changing opportunity costs over the VC fund life-cycle”. Under ILPA’s model agreement, capital called after the commitment period can pay expenses, fund follow-ons or close deals already signed, and nothing else. Late in a fund, a reserved dollar has no new company to compete with.

Isin Guler found the same reluctance to stop. She examined US venture investments from 1989 to 2004. Firms “become less likely to terminate investments as they participate in more rounds of financing, despite evidence that expected returns are declining over rounds”, with internal politics and pressure from co-investors and limited partners among the likely influences (Guler, 2007).

Students of animal behaviour call the error the Concorde fallacy, after the British and French governments that kept funding the airliner “even after it became apparent that there was no longer an economic case for the aircraft” (Wikipedia, Sunk cost). The two governments were bound by a treaty; each investor in a syndicate is free to stop, which makes the insiders’ persistence harder to excuse. Hal Arkes and Peter Ayton found “no unambiguous instances of the Concorde fallacy in lower animals” (Arkes & Ayton, 1999).

The exposure is growing. On Carta’s data, 30.6% of companies that raised a seed round in Q1 2018 reached a Series A within two years, against 15.4% of the 2022 cohort (Carta, 2024). With half as many outside prices, more of every reserve meets requests only insiders will price.

But the fund’s own terms reward holding reserves, spent or not

A reserve changes the manager’s economics before a dollar of it is spent. ILPA’s model agreement lets the manager charge fees on a successor fund once 80% of commitments are “funded, invested, committed or reserved for investments (including Follow-on Investments)”. A dollar set aside counts toward the next fund as fully as a dollar invested.

Fees can follow the reserve too. A 2023 staff memo for Rhode Island’s state pension shows that GGV Capital IX, after its investment period, charges 2.5% a year on “the sum of invested capital and uncalled capital reserved for investments” until year 10 (Rhode Island Treasury, 2023). The same memo shows another design: GGV IX Plus, a sidecar that follows on in “high potential GGV IX portfolio companies” with no management fee, took $1.6M from Rhode Island for every $6.4M committed to the main fund.

Carried interest points the same way. In Sapphire’s worked example, if reserve dollars return 2x, the manager’s carry rises from $44M to $55M while an LP’s multiple falls from 4.0x to 2.7x. None of this implies bad faith (Rhode Island’s staff judged GGV’s fees in line with industry standards), but it makes the reserve’s size an economic term, to be negotiated like a fee.

Therefore size the reserve from the rounds outsiders will price

If the outside price carries most of the information, the reserve that matters buys pro rata in priced rounds, and its size can be estimated. At Carta’s 2026 medians, an investor that keeps its stake pays about $0.53 at the Series A and $1.07 at the Series B per $1 of seed check.

Exhibit 3. What pro rata in priced rounds costs, per $1 of first checks

	Series A	Series B	Total
Pro rata per $1 of seed check, to keep the stake	$0.53	$1.07	$1.59
Seed companies reaching the round	30.6% (2018 cohort) or 15.4% (2022 cohort)	Half of those (hypothetical)
Priced pool per $1 of first checks	$0.16 or $0.08	$0.16 or $0.08	$0.32 or $0.16
Priced pool as a share of investable capital			24% or 14%

Source: Carta (2026) median post-money valuations ($21.6M seed, $63M Series A, $177M Series B) and median dilution (18% at Series A, 13% at Series B); Carta (2024) for graduation within two years. The Series B rate is hypothetical. All figures are our arithmetic.

On these numbers, pro rata in priced rounds needs $0.16 to $0.32 per dollar of first checks, a reserve of 14% to 24% of investable capital. A 1:1 plan holds $1, so $0.68 to $0.84 of every reserve dollar goes to super pro rata, later rounds, bridges and insider rounds, or stays uncalled. Late graduates raise the first figure, so treat it as a floor.

The contract makes this priced pool the one a fund can count on. Under the NVCA’s model investors’ rights agreement, the company must first offer new securities to each Major Investor, a status set by a share threshold the parties fill in (NVCA, 2025). Each has 20 days to elect its pro rata share. An optional clause ends the right for a Major Investor that fails to buy its full pro rata amount, so one skipped round can cost every later one.

Therefore test the rule: in a worked model, reserves trade the typical fund for a fatter tail

The simulation below varies the reserve ratio, the share of companies that get follow-on money and the partners’ skill, on these assumptions:

• A $100M fund with $80M to invest after 20% of fees, near the 17.75% median Andrew Metrick and Ayako Yasuda found (Metrick & Yasuda, 2010).
• $2M first checks, with outcomes drawn from the power law Moonfire Ventures fitted to Correlation Ventures, AngelList and Horsley Bridge data, wins capped at 100x (Farina et al., 2023).
• A Series A price 2.39 times the seed price (Carta).
• 60% of eventual winners and 16% of the rest raise a priced Series A, about 30.5% overall against Carta’s 30.6% (our assumption).
• Follow-on checks capped at twice the first check, Wilson’s ceiling. Skill is how often the partners rank the better of two graduates first; chance is 50%.

Exhibit 4. Reserve planning: VC fund outcomes under six rules, 10,000 simulated funds each

Reserve and follow-on rule	First checks	Mean multiple	Median multiple	Funds at 3x or more	Funds below 1x	Reserve deployed
No reserve	40	1.89x	1.53x	16%	18%	n/a
25%, every graduate	30	1.79x	1.36x	15%	28%	99%
50%, every graduate	20	1.70x	1.17x	13%	39%	61%
50%, top third, picked at chance	20	1.79x	1.24x	14%	35%	24%
50%, top third, right 3 times in 4	20	2.01x	1.35x	17%	30%	24%
50%, every graduate, rest in insider rounds	20	1.62x	1.12x	12%	42%	100%

Source: CX Cash model, code below. Gross multiples on capital invested, before carry; follow-ons all go in at the Series A price, which flatters them. All figures are our arithmetic under the stated assumptions.

import math, random, statistics as st

ALPHA, XMIN, XMAX = 2.05, 0.35, 100.0  # power law fitted by Moonfire (2023); 100x cap is ours
STEP_UP = 2.39                         # Series A price / seed price (Carta Q1 2026 medians)
GRAD = {True: 0.60, False: 0.16}       # P(priced Series A | outcome >= 1x, < 1x): our assumption

def outcome(rng):                      # gross multiple on a seed dollar
    x = XMIN * (1 - rng.random()) ** (-1 / (ALPHA - 1))
    return min((x - XMIN) / (1 - XMIN) if x < 1 else x, XMAX)

def fund(rng, reserve, share=1.0, skill=0.0, insiders=False, money=80.0, check=2.0, cap=4.0):
    n = round(money * (1 - reserve) / check)               # seed checks the fund can write
    m = [outcome(rng) for _ in range(n)]
    grads = [i for i in range(n) if rng.random() < GRAD[m[i] >= 1]]
    value, paid, left = check * sum(m), check * n, money * reserve
    rank = {i: r / len(grads) for r, i in enumerate(sorted(grads, key=m.__getitem__), 1)}
    picks = sorted(grads, key=lambda i: -(skill * rank[i] + (1 - skill) * rng.random()))
    picks = picks[:math.ceil(share * len(grads))]
    for pool, price in ((picks, STEP_UP), ([i for i in range(n) if i not in grads] if insiders else [], 1.0)):
        if pool and left > 0:                              # priced rounds first, then insider rounds
            each = min(cap, left / len(pool))
            value += sum(each * m[i] / price for i in pool)
            paid, left = paid + each * len(pool), left - each * len(pool)
    return value / paid, (1 - left / (money * reserve)) if reserve else 0.0

def row(reserve, share=1.0, skill=0.0, insiders=False, sims=10_000):
    rng = random.Random(2026)
    runs = [fund(rng, reserve, share, skill, insiders) for _ in range(sims)]
    x = [a for a, _ in runs]
    return (round(80 * (1 - reserve) / 2), st.mean(x), st.median(x), sum(v >= 3 for v in x) / sims,
            sum(v < 1 for v in x) / sims, st.mean(b for _, b in runs))

print(row(0.5, share=1/3, skill=0.5))   # the fifth row of Exhibit 4

In this model, every reserve lowers the median fund, as following on did in AngelList’s deals, and raises the share of funds that lose money. The mean beats the no-reserve fund only when the partners pick the better graduate three times in four, against 50% for chance. At Carta’s graduation rate a 50% reserve cannot all reach priced rounds (61% is deployed when the fund follows every graduate, 24% when it concentrates), and spending the rest in insider rounds gives the worst row.

The model turns on the step-up test. Its graduates are expected to return 1.80 times the average seed outcome, short of the 2.39 step-up, so following all of them dilutes the fund. A sharper graduation signal lifts the mean to about level with no reserve, and the median still falls. Moonfire’s simulations agree: “a portfolio with 50 investments and no follow-on has the same risk of a portfolio of 90 investments with 20% allocated to follow-ons”.

Wilson’s objection holds for the winners’ priced rounds

Fred Wilson states the case for large reserves: “One of the most common mistakes I see new ‘emerging VC managers’ make is that they don’t sufficiently reserve for follow-on investments” (Wilson, 2017). USV models every future round and holds enough for “a 95% probability of being able to participate in all of these future funding rounds”. In Sapphire’s funds that returned 5x or more, the top company averaged about 90x, the second about 25x and the rest about 1x. In Sapphire’s example, a fund that follows on in every Series A but misses the outlier falls from 4.0x to 2.5x net. And AngelList’s always-follow policy had the higher mean, which is what an LP spread across many funds collects.

Wilson is right about priced rounds in the winners, the part of reserve planning VC managers can least afford to get wrong. A fund that cannot take its pro rata in its best company’s Series B loses ownership in the likeliest fund-returner, and Wilson’s per-company ceiling is modest: “as much as $6mm (2x the initial investment) reserved for it. Don’t expect more than that.”

His case does not reach the rest of a 1:1 reserve. Money spent where no outsider will price is the money most likely lost. An LP in a single fund lives with that fund’s median. Hunter Walk, a co-founder of Homebrew, reversed himself in July 2026 for funds of $100M or less, urging minimal reserves “so that you are not thinking about it as a second pool of dollars to only use for second checks” (Walk, 2026).

Two pools for partners, four questions for LPs

• Pools. Split the reserve into a priced pool, sized from graduation and pro rata (Exhibit 3), and a capped conviction pool spent only after a written step-up test.
• Insider rounds. Treat a round only insiders fund as a new investment the fund prices itself; compare it with a new first check.
• Releases. Release a company’s reserve once it will not graduate, and say so in the quarterly letter.
• Rights. Check investors’ rights agreements for the Major Investor threshold and the lapse clause.
• LP returns. Ask for follow-on multiples split by who set the price, and the graduation rate and step-up behind the reserve model.
• LP terms. Check whether reserved capital counts toward the successor-fund trigger (80% in ILPA’s model) and whether later fees run on reserved, uncalled capital.
• LP caps. Negotiate the caps on late follow-ons (ILPA’s model brackets 15% within 18 months) and on recycling (some LPs use 110% of commitments).
• LP sidecars. Prefer a fee-free sidecar for large follow-ons, as Rhode Island did with GGV IX Plus.

Size for priced rounds and release the rest

A follow-on buys the same company at about 2.4 times the seed price, and the partners’ information beats an outside lead’s price only modestly. Where a new investor sets the price, following on has roughly held value; where only insiders fund the round, returns fall. Fund terms reward holding the reserve either way. In reserve planning, VC funds can therefore size the reserve for priced rounds, cap the rest and release what no outsider will price. One input is still moving: Carta’s two-year graduation rate fell by about half between the 2018 and 2022 seed cohorts, and if it stays near 15.4%, the priced pool shrinks to about 14% of investable capital.

Sources and notes

• Arkes, H. R. and Ayton, P. (1999). The sunk cost and Concorde effects: are humans less rational than lower animals? Psychological Bulletin 125(5), 591–600.
• Carta (2024). VC Fund Performance: Q1 2024 (P. Walker, M. Young, K. Dowd and A. Lester).
• Carta (2026). State of Private Markets: Q1 2026, full report.
• Ewens, M., Nanda, R. and Rhodes-Kropf, M. (2018). Cost of experimentation and the evolution of venture capital. Journal of Financial Economics 128(3), 422–442; NBER Working Paper 24523.
• Ewens, M., Rhodes-Kropf, M. and Strebulaev, I. A. (2016). Insider financing and venture capital returns. Stanford GSB Research Paper 16-45, SSRN 2849681.
• Farina, F., Arpaia, M., Khing, H. and Vetterle, J. (2023). Venture capital portfolio construction and the main factors impacting the optimal strategy. Moonfire Ventures, arXiv:2303.11013.
• Gompers, P. A. (1995). Optimal investment, monitoring, and the staging of venture capital. The Journal of Finance 50(5), 1461–1489.
• Guler, I. (2007). Throwing good money after bad? Political and institutional influences on sequential decision making in the venture capital industry. Administrative Science Quarterly 52(2), 248–285.
• Institutional Limited Partners Association (2020). Model Limited Partnership Agreement (whole-of-fund), July.
• Knicker, M. S., Bouchaud, J.-P. and Benzaquen, M. (2026). Do venture capitalists beat random allocation? arXiv:2605.03980.
• Metrick, A. and Yasuda, A. (2010). The economics of private equity funds. Review of Financial Studies 23(6).
• National Venture Capital Association (2025). Model Investors’ Rights Agreement, revised October 2025.
• Othman, A. (2019). Should Seed Investors Follow On? AngelList, 1 September.
• Rhode Island Office of the General Treasurer (2023). GGV Capital IX, L.P., GGV Capital IX Plus, L.P., and GGV Discovery IV-US, L.P.: Staff Recommendation, February.
• Sapphire Partners (2022). Dirty Secret: Venture Reserves Are Not Always a Good Thing. Sapphire Ventures blog, 4 May.
• The Fund CFO (2026). Rethinking VC reserves.
• Walk, H. (2026). I’ve changed my mind: early stage venture funds of $100 million or less should hold almost no reserves for follow-on. 23 July.
• Wikipedia. Sunk cost.
• Wilson, F. (2017). Reserves. AVC, January.

The figures come from the papers, fund documents and market data listed above; the AngelList, Carta and Sapphire numbers are vendor or LP data, and the Knicker paper is a working paper. The findings of Ewens, Rhodes-Kropf and Strebulaev, Guler, Gompers, and Arkes and Ayton are cited from their abstracts. Figures that cite no source are derived from the cited data, and Exhibit 4 is a simulation under the stated assumptions, which the code above reproduces with the standard Python library.