Abstract
This paper makes a quantitative case for monthly-resolution cash-flow modeling in LP commitment pacing, against the still-common practice of annual or even quarterly resolution. The argument is structural rather than rhetorical: capital calls and distributions in private markets are not evenly distributed across the year, and a model that smooths them across twelve months systematically understates the peak-to-trough swing in unfunded commitment. The peak-to-trough swing is the figure that determines how large a liquidity reserve a fiduciary needs to hold; underestimating it is a real fiduciary exposure, not a modeling preference.
We illustrate the argument with a worked synthetic 6-fund portfolio across three asset classes, comparing the unfunded-commitment trajectory at annual vs. monthly resolution. The peak-unfunded gap at the trough of the annual model can run 15–25% below the actual peak observed under monthly resolution — a gap that translates directly into under-sized liquidity buffers. We then walk through where the lumpiness comes from per asset class, what monthly-resolution implementation actually requires (data model, engine, re-run cadence), and provide a six-question diagnostic for institutions evaluating their own pacing models.
The paper is honest about implementation state. Meridian's underlying fund simulators run at monthly resolution; the live-commitment forecast service that drives operational dashboards currently produces quarterly projections, with a monthly extension on the roadmap. Per-asset-class calibration depth varies — buyout and private credit have the most refined calibration; real estate, infrastructure, venture capital, and natural resources have asset-class-specific deterministic models with calibration refinement in active progress. The framework described here is what every institutional LP pacing model should aspire to; the implementation gap is named explicitly throughout.
The audience for this paper spans both sides of an institutional LP. On the operations side: treasurers and operations leads sizing liquidity buffers against projected unfunded commitment, and audit-committee members for whom pacing-model defensibility is part of program governance. On the investment side: CIOs and investment-committee members making allocation decisions whose downstream cash-flow shape this paper describes. The discipline matters in both contexts, for related but distinct reasons.
1. The Pacing Problem in Resolution Terms
Commitment pacing — the schedule and sizing of new fund commitments over time — is one of the most consequential decisions an LP makes. A pension fund that commits $100M to a buyout fund in 2024 will see that capital deployed over 4–5 years and returned over 8–12 years. The pacing decisions made today shape portfolio composition, cash-flow profile, and liquidity exposure for well over a decade.
What is less commonly recognized is that the resolution at which pacing is modeled is itself a fiduciary decision. A pacing model that produces an annual unfunded-commitment trajectory and a pacing model that produces a monthly trajectory are not the same model with different reporting granularity. They are different models. They produce different peaks, different troughs, different stress-scenario behavior, and — most consequentially — different liquidity-buffer recommendations.
The choice between annual, quarterly, and monthly resolution is therefore not a presentation choice. It is a model choice that affects what the fiduciary sees and what the fiduciary plans against. A pacing model that smooths lumpy cash flows across twelve months will report a smoother, more comfortable trajectory than the underlying portfolio actually experiences. A treasurer relying on that smoothed trajectory to size the liquidity reserve will under-size it. The error is structural, not a calibration adjustment.
This paper is about why that happens, what the magnitude looks like, and what the discipline looks like to fix it.
2. Where Annual Pacing Models Fail
Annual pacing models — by which we mean any model whose output is one data point per year per metric — fail in three specific, measurable ways.
Peak-to-trough understatement. Capital calls and distributions are not uniformly distributed across the year. Capital calls cluster around quarter-end (when GPs close on portfolio company acquisitions) and around vintage anniversaries (when initial drawdowns happen). Distributions cluster around realized exits, which themselves cluster around year-end (when GPs push to close transactions before year-end reporting cutoffs) and around IPO windows. An annual model averaging these into smooth twelve-month flows will report an unfunded balance that is the average of the intra-year fluctuation, not the peak. The peak is what matters for liquidity sizing. The average is what gets reported. The two diverge by a structurally meaningful amount.
Liquidity-buffer mis-timing. The liquidity reserve an institution holds against its private-markets program is sized to cover the worst-case capital-call demand the portfolio could produce within the institution's response window. A treasurer who sees an annual model showing peak unfunded of $90M will plan a buffer against $90M. A treasurer who sees a monthly model showing the same year's peak unfunded reaching $112M during the third quarter — before declining as Q4 distributions arrive — plans against $112M. The difference is not academic; the institution that sized to $90M is exposed when the third-quarter call wave arrives and the buffer is insufficient.
Mid-year reconciliation drift. When actual cash flows arrive, the institution reconciles them against the pacing model's projection. With annual resolution, the comparison is "annual actual vs. annual projected" — a single data point per year. Variances within the year cancel out and never surface as a model-quality signal. With monthly resolution, the model is testable twelve times per year. Months where the model materially mis-predicts become diagnostic: they identify which assumption (call timing, deployment pace, distribution schedule) is the source of error, and they accumulate into a calibration record over time. An annual model is functionally untestable at any resolution finer than the year itself.
The third failure mode is the most consequential for long-term model quality. A model that cannot be tested cannot be improved. Institutions that have run annual pacing models for a decade have run them without ever exposing them to the kind of mid-year scrutiny that would surface their structural biases. The biases compound; the models become more confident than they should be; the gap between projected and actual cash-flow behavior is invisible to the institution running the model.
3. The J-Curve as a Quarterly-and-Monthly Phenomenon
The J-curve is the standard term for the shape of cumulative net cash flow in a private fund: contributions outpace distributions in early years (NAV is below contributed capital), the curve crosses zero somewhere mid-life (often years 4–6 for buyout funds, later for venture and infrastructure), then distributions outpace contributions in later years as exits realize.
What the textbook depiction of the J-curve smooths over is that the underlying mechanism — capital calls and distributions — is not smooth. A buyout fund's capital calls do not arrive evenly. They arrive in clumps: an initial drawdown at first close, then quarterly drawdowns of varying size as the GP closes on portfolio companies, then a final drawdown phase when the fund commits its tail capital. A real estate fund's calls cluster around acquisition closings (which themselves depend on transaction-market conditions). An infrastructure fund's calls are heavily back-loaded toward construction-phase capital expenditure, which can compress into 6–12 month windows when major projects break ground.
Distributions are similarly lumpy. A buyout fund's distributions arrive when portfolio companies are sold; a year with three exits looks materially different from a year with zero. A real estate fund's distributions arrive at refinancings and dispositions. An infrastructure fund's distributions begin as ongoing cash yield (predictable) but step up sharply at exit (lumpy).
Annual pacing models smooth all of this. They take a fund's expected lifetime call profile (say, 25% in year 1, 20% in year 2, 20% in year 3, 15% in year 4, etc.) and treat each year's call as if it arrived evenly across twelve months. The resulting unfunded-balance projection is a piecewise-linear curve — a series of straight lines connecting year-end snapshots. The actual underlying behavior is a step function with sharp jumps at irregular intervals.
The mathematical consequence: the annual model's reported peak unfunded is the maximum of a piecewise-linear curve passing through annual snapshots. The actual peak unfunded is the maximum of a step function whose jumps are concentrated. The actual peak is always greater than or equal to the annual model's reported peak — because the actual has higher temporal frequency, and thus higher local variance, around any underlying smooth trend.
The gap between the two depends on the lumpiness of the underlying flows. For asset classes with highly clustered cash flows (real estate at acquisition closings, infrastructure at construction phase, late-stage venture at follow-on rounds), the gap can be 15–25% of peak unfunded. For asset classes with smoother flows (buyout in mature deployment, private credit with regular coupon distributions), the gap is smaller — perhaps 5–10%. In either case, the gap is structural; it does not go away with better calibration of the annual model. Only finer temporal resolution removes it.
4. A Worked Illustration
Consider a synthetic 6-fund portfolio held by a hypothetical institutional LP with $1.5B in total private-markets commitments. The portfolio is composed as follows:
| Fund | Asset class | Vintage | Commitment | Years to first close from now |
|---|---|---|---|---|
| Buyout Fund A | Buyout | 2023 | $300M | -2 (mid-deployment) |
| Buyout Fund B | Buyout | 2025 | $250M | 0 (just closed) |
| Real Estate Fund A | Real Estate | 2024 | $200M | -1 (early deployment) |
| Infrastructure Fund A | Infrastructure | 2024 | $300M | -1 (entering construction phase) |
| Private Credit Fund A | Private Credit | 2025 | $250M | 0 (just closed) |
| Venture Fund A | Venture | 2026 | $200M | +1 (committed; first close pending) |
We project the portfolio's unfunded commitment over the next 36 months under two models: an annual-resolution model that interpolates smoothly between year-end snapshots, and a monthly-resolution model that respects the underlying call/distribution timing per asset class.
Annual model report:
| Year-end | Reported unfunded |
|---|---|
| Year 1 | $612M |
| Year 2 | $478M |
| Year 3 | $341M |
The annual model reports a smoothly declining unfunded balance, peaking at $612M at the end of Year 1.
Monthly model report (key months):
| Month | Actual unfunded | Notes |
|---|---|---|
| Month 6 (Year 1) | $681M | Infrastructure construction-phase calls + Venture first close |
| Month 12 (Year 1) | $612M | Year-end snapshot — same as annual model |
| Month 18 (Year 2) | $574M | Real Estate acquisition close + Private Credit drawdown |
| Month 24 (Year 2) | $478M | Year-end snapshot — same as annual model |
| Month 30 (Year 3) | $402M | Buyout B mid-deployment |
| Month 36 (Year 3) | $341M | Year-end snapshot — same as annual model |
The two models agree at year-end — they have to, because they are both calibrated to the same fund-level commitment schedules. The two models disagree mid-year. Specifically, the monthly model surfaces three mid-year peaks (Month 6, Month 18, Month 30) that the annual model entirely misses. The Month-6 peak of $681M is 11.3% above the annual model's reported Year-1 peak of $612M. A treasurer relying on the annual model would size a buffer for $612M; the actual portfolio touches $681M six months in.
The downstream consequences:
- The buffer needs to be ~$70M larger than the annual model would suggest, on a $1.5B program. That's the difference between adequate and inadequate liquidity coverage.
- The mid-year peaks identify which funds' call concentrations are creating the pressure. The Infrastructure construction-phase ramp at Month 6 is a known event for that asset class; an institution that monitors monthly can pre-position cash for it. An institution that sees only annual snapshots cannot.
- The model becomes testable monthly. Each month's actual unfunded vs. projected unfunded is a data point; over a year, twelve data points provide a calibration signal that an annual model produces only one of.
A note on the illustration: the synthetic portfolio above is constructed to demonstrate the gap mechanism. The exact magnitudes will vary by portfolio composition. A portfolio heavier in private credit (smoother flows) and lighter in infrastructure (lumpier flows) will see a smaller gap; a portfolio heavier in real estate and infrastructure will see a larger one. The qualitative pattern — annual under-reports the peak, monthly surfaces it — is structural and applies to every multi-asset-class portfolio.
5. Per-Asset-Class Behavior
The lumpiness mechanism varies materially by asset class. A pacing model that uses a single set of call/distribution assumptions across all asset classes will produce particularly bad results for the asset classes whose actual behavior diverges most from the average.
Buyout. Capital calls cluster around portfolio-company acquisitions, which themselves cluster around quarter-ends and around favorable transaction-market windows. A typical mid-market buyout fund calls 60–70% of commitments in years 2–4, with intra-year concentration in 2–3 acquisition windows per year. Distributions arrive at exits, typically beginning year 4–5 and accelerating through year 7–9. The intra-year clustering is moderate — calls and distributions are lumpy but not extreme.
Private Credit. Capital calls front-load heavily, often 50–60% in the first two years as the fund builds the loan book. Once deployed, the portfolio generates regular cash yield — quarterly or semi-annual coupon distributions — that are predictable in timing. End-of-life distributions arrive when loans mature or refinance. Intra-year clustering is low on the distribution side (yield is regular) and moderate on the call side (concentrated in the deployment phase).
Real Estate. Capital calls cluster around acquisition closings, which depend on transaction-market timing and individual deal-specific timelines. A real estate fund with three large acquisitions in a year will call most of its annual commitment in three discrete windows. Distributions arrive at refinancings (sporadic) and dispositions (clustered toward fund end-of-life). Intra-year clustering is high.
Infrastructure. Capital calls back-load toward construction phase, which can compress 30–40% of total commitments into 6–12 month windows when major projects break ground. Distributions begin as ongoing yield once assets are operational (predictable) and step up sharply at exit (lumpy). Intra-year clustering during construction phase is very high.
Venture Capital. Capital calls follow follow-on round timing, which depends on portfolio-company stage and fundraising-market conditions. Early years are typically light on calls (the GP is making initial investments); later years see follow-on calls clustered around portfolio companies' Series rounds. Distributions are extremely lumpy — concentrated entirely on the small number of exits that materialize. Intra-year clustering is high on both sides.
Natural Resources. Cash flows depend heavily on commodity-price cycles and on project-specific operational timing (mine development, well drilling, harvest cycles). Intra-year clustering varies by sub-strategy but is generally high.
A pacing model that uses a single intra-year call profile across all six asset classes will systematically over-smooth Real Estate, Infrastructure, and Venture (high clustering), and under-smooth Buyout and Private Credit (moderate-to-low clustering). The portfolio-level error compounds across asset classes; a multi-asset portfolio's true peak unfunded reflects the worst clustering of its component asset classes, not the average.
Implementation note: Meridian's fund simulators are asset-class-specific. Buyout and Private Credit have the most refined calibration — the underlying engine logic was ported from a longer-running web2py simulator with multi-cycle calibration data. The Real Estate, Infrastructure, Venture Capital, Natural Resources, Growth Equity, Fund-of-Funds, and Secondaries simulators are deterministic asset-class-specific models that capture the qualitative clustering pattern described above (e.g., infrastructure construction-phase concentration, venture follow-on timing) but with simpler parameter calibration than Buyout/Private Credit. Per-asset-class refinement work — adding probabilistic dispersion bands, calibrating against per-fund historical schedules, surfacing confidence intervals — is in active progress. The framework described in this paper is what every asset-class simulator should aspire to; the implementation gap varies by asset class and is being closed iteratively.
6. The Implementation Discipline
Monthly-resolution pacing is not just a finer-grained version of the same model. It is a different implementation discipline.
Data model. A monthly pacing model needs a data structure that can hold per-month projected and actual cash flows for each fund, and aggregate across funds at any point in time. The annual-model approach of storing one number per fund per year is not sufficient; the monthly model needs twelve numbers per fund per year, with the ability to compare projected vs. actual at each month. The data model is therefore richer and the storage cost is non-trivially higher — but the storage cost is a rounding error compared to the cost of a mis-sized liquidity buffer in a stressed quarter.
Engine architecture. The pacing engine itself must be capable of producing month-by-month projections. The engine needs per-asset-class call and distribution profiles at monthly resolution, not annual aggregates. The simulation must produce a step-function trajectory rather than a piecewise-linear interpolation. And the engine must be re-runnable — every month, against the latest actual cash-flow data, with the resulting projection updating the forward view.
Re-run cadence. A monthly model is only valuable if it is re-run monthly. An institution that builds a monthly engine and re-runs it annually has a monthly engine producing annual reports. The re-run cadence is therefore a discipline question: is the institution committing to monthly operational cadence, or is the model going to drift quickly toward stale projections?
The practical re-run triggers should include:
- Scheduled monthly run at month-end against the latest actual cash flows
- Event-driven re-run when a new commitment closes (the new fund's call profile must be added to the projection)
- Event-driven re-run when an IC-approved policy-band change occurs (the operational ceilings the projection is tested against will have moved)
- Event-driven re-run when material assumptions change (e.g., when a GP signals an acceleration in deployment pace)
Each re-run is versioned; prior runs are not deleted. The trajectory of forecast revisions over time becomes itself an audit-trail artifact — available to the IC if a forecast turns out to have meaningfully diverged from realized outcomes.
Reporting discipline. Monthly resolution at the engine level should not necessarily mean monthly resolution in every report. A board package may still report annually because that is the cadence at which the board operates. But the annual report should be a projection of the monthly model, not a separate annual model. The annual numbers should be derivable from the monthly numbers; when the monthly numbers update, the annual numbers update; the annual report is a view, not an alternative source of truth.
Implementation note: Meridian's underlying fund simulators (
buyout_fund_simulator,credit_fund_simulator, plus the asset-class-specific simulators for the other classes) operate at monthly resolution. The live-commitment cash-flow forecast service that drives operational dashboards (cash_flow_forecast_service,pacing_models.py) currently produces quarterly projections, not monthly. The framework described above implies extending the live-forecast service to monthly resolution; this is on the engineering roadmap but not yet shipped. The honest current state: monthly engine, quarterly live-operational forecast, monthly extension in progress.
7. A Diagnostic for Your Current Pacing Model
The following six questions test whether an institution's current pacing model is genuinely monthly-resolution or is annual-disguised-as-monthly:
-
Does the model produce a per-month projected unfunded balance for the next 24+ months? If the answer is "we report annually but the engine could produce monthly if we asked," the engine is annual; ask the engineer running it whether the call/distribution assumptions are stored at monthly granularity or as annual aggregates.
-
When you compare projected vs. actual at the end of a quarter, do you get one variance number for the quarter or three (one per month)? If one, the model is producing annual-equivalent output regardless of its label. If three, the model is genuinely tracking month-by-month.
-
Does the model differentiate intra-year call timing across asset classes? A model that uses a single call profile for all asset classes is structurally averaging; one that uses asset-class-specific profiles is more honest about the lumpiness.
-
When a new commitment closes mid-year, does the model immediately update the next 12+ months of projections, or does the change wait until the next annual cycle? The first is operational; the second is annual-disguised-as-monthly.
-
Does the institution's liquidity-reserve sizing reference the model's peak unfunded across the full horizon, or its year-end unfunded snapshots? The first is the right answer; the second is the mis-sized-buffer trap.
-
Can the model reproduce, for any month in the trailing year, what the projected balance was for that month as forecast 12 months prior? If yes, the institution has versioned monthly projections — a real audit trail. If no, the model is not preserving its own forecast history and cannot be evaluated for calibration drift.
A model that answers "yes" to all six is genuinely operating at monthly resolution. A model that answers "no" to two or more is producing reports at monthly granularity but is annual or quarterly underneath — the operational behavior the institution actually has is the looser one.
8. What This Means
The case for monthly-resolution pacing is not aesthetic. It is structural and quantitative. Annual models systematically understate peak unfunded commitment in ways that produce mis-sized liquidity buffers, and they hide model-calibration errors that monthly models would surface. Both consequences accumulate over time and matter most exactly when they matter most — during stressed quarters where actual capital-call demand exceeds the buffer the institution sized against an annualized projection.
The implementation discipline required to operate at monthly resolution is real but not exceptional. A data model that stores per-month projected and actual cash flows; an engine that produces step-function trajectories per asset class; a re-run cadence that includes scheduled monthly runs and event-driven re-runs at commitment closings, policy-band changes, and material assumption updates; and a reporting layer that derives annual views from the monthly model rather than maintaining a separate annual one. Each piece is a defined engineering and operational commitment; each can be evaluated independently; together they constitute a pacing capability that survives the kind of scrutiny a fiduciary review will eventually apply.
The category-wide state of practice is a long way from this standard. Most institutional pacing tools — including the spreadsheets that many institutions still use — operate at annual resolution, sometimes with quarterly snapshots layered on top. The transition to monthly is genuine engineering work. Institutions evaluating their own current pacing infrastructure should apply the diagnostic in §7 honestly; an institution that answers "no" to several questions is operating with a real fiduciary exposure that the institution may not be aware of.
The framework described in this paper is what the destination looks like. The implementation gap — including Meridian's gap, where the underlying simulators are monthly but the live-commitment forecast service is quarterly — is named explicitly so that institutions evaluating the platform can evaluate the gap honestly. The intent is to set the standard, describe the discipline required to meet it, and be candid about which pieces of the standard are shipped today and which are in progress.
A pacing model that under-sizes a liquidity buffer is not a modeling preference. It is a fiduciary exposure that will surface — invisibly, then suddenly — the next time a stressed quarter compresses the call timeline an annual model has been smoothing.
This paper completes the methodology trilogy on fiduciary discipline in institutional LP technology. See also: ILPA Reconciliation: Signed-Convention Formulas for LP Auditability (data discipline) and Multi-Channel GP Research as Fiduciary Practice (AI/inference discipline).