Reconstructed Rebalancing Ledgers: A Transparency Standard for Simulated Index Products

1. Why rebalancing histories matter

The reference practice is a disclosure chain, not a single document. S&P Dow Jones Indices announces index changes roughly five trading days before they take effect, with pro-forma constituent files, and applies them on a fixed public calendar; MSCI's reviews take effect on the first of the following month; FTSE Russell runs the same quarterly pattern. The major ETF issuers publish holdings daily, so any tracker's realised turnover is auditable by subtraction. The chain exists because turnover is a priced quantity: every percentage point of one-way turnover costs the holder roughly twice its execution spread, and an index that will not show its rebalancing record is asking to be trusted about a number that debits returns directly.

A simulated product sits outside this chain by construction. Nothing executes, so there are no fills to audit — but the published statistics embed rebalancing assumptions all the same, and always did. Every Sharpe ratio in the OLTA catalogue is produced by a simulation that resets weights on some cadence; the cadence, its boundary semantics, and the turnover they generate are real properties of the number, yet until this program they were visible only as a one-word label next to the result. A simulated product, having no execution record at all, needs the reconstruction most, not least.

The standard this paper documents is short: every published performance figure should be traceable to a published event history under a stated convention. The rebalancing program treats three rules as non-negotiable. First, numbers follow the simulation — any cadence named beside a statistic is the cadence that produced it. Second, policy may differ from simulation, and where it does the product says so once, plainly, per index. Third, every event timeline, turnover figure, and drift readout carries the reconstruction disclosure: simulated, never executed.

2. Methodology: declared-cadence reconstruction

The ledger reconstructs each index at its declared cadence — the rebalancing policy the catalogue states for the product, with the default (quarterly) recorded as a default rather than passed off as a declaration. This is a deliberate choice of question. The reconstruction answers "what would the stated rulebook have done over this window?", which is the question a rebalancing surface owes its reader; whether the published statistics were simulated under that same rulebook is a separate question the ledger answers separately (Section 4).

The simulation layer is the engine already documented in the methodology paper: divisor-method NAV in the S&P convention, with share counts recomputed at each rebalance and the divisor calibrated once at inception; eight rebalance strategies (buy-and-hold, weekly, monthly, quarterly, drift-threshold, and three strategic overlays); calendar boundaries firing on the first calendar day of each period, UTC. The engine's rebalance events natively carry the pre- and post-event weight vectors; NAV at each event joins from the same series the statistics are computed on. Data preparation is a single implementation shared with the analysis pipeline (history-window selection, venue merge, stablecoin synthesis, availability filtering with weight rescale), so ledger and statistics cannot drift apart through duplicated preparation code. Each file stamps its coverage: drift is measured against the post-coverage rescaled target vector the simulation actually ran on, and 27 of the 28 files sit at 100% symbol coverage (OBTCA50 at 98%).

Each event records the date and boundary type, the drifted pre-event weight vector, the post-event target vector, one-way turnover defined as half the sum of absolute weight changes, the engine-reported drift at trigger, and NAV at the event. Event lists are complete — uncapped, newest first, with the day-zero initial allocation included as a labelled synthetic event carrying no turnover. The summary block reports event count, average per-event and annualised one-way turnover, and a cost-sensitivity table at 5, 10, and 20 basis points under a convention printed verbatim in every file: annual return drag equals two times one-way annualised turnover times the cost assumption — both legs of the traded notional, no market impact modelled. A drift status block reports current weights against target on a single labelled basis — since the last simulated rebalance — so a drift figure is never compared against a basis other than the one printed next to it.

Selection is the tradable book: 28 of the 29 tradable indices in the 84-index catalogue. The 29th, OAH7, declares a session cadence the engine has no strategy for; it is skipped with that reason recorded in the summary file rather than reconstructed under a cadence it does not declare. Retired indices carry no rebalancing surface at all — a retired product with a live-looking event timeline would be its own small dishonesty. Every file closes with a provenance block — simulated true, basis "backtest-reconstruction", generator identity, generation timestamp, data as-of date, history window — so a reader who loads any ledger file without this paper still gets the disclosure.

3. What the ledger shows

The July 23, 2026 generation covers 12 quarterly-declared indices, 12 monthly, and 4 drift-triggered, over windows from 106 days (the youngest basket) to 1,824 days (the five-year equity baskets). Twenty-four indices produced 533 reconstructed events between them; the four drift-triggered indices produced none.

Two regularities stand out. Cadence dominates family: the monthly cohort clusters between 36% and 64% a year whether the basket is crypto, equity, or cross-asset, while quarterly baskets span an order of magnitude — from 5.6% (a five-name broad-ETF basket whose constituents barely disperse) to 76.4% (a three-name basket whose constituents disperse violently). And per-event turnover is a concentration-and-dispersion story: average one-way turnover per event runs from 1.4% on the quietest basket to 17.3% on the loudest, with basket size and constituent volatility, not cadence, setting the figure. Every index carries less than 0.16% a year of implied drag at 10 basis points and less than 0.31% at 20 — modest figures, but now published per index from a stated formula rather than asserted.

OCI10, the routine case. Core 10, quarterly, 715-day window from 2024-08-08, eight reconstructed events, average 8.1% one-way per event, 33.0% annualised. The January 1, 2025 event is the reconstruction at its most informative: the Q4-2024 repricing had carried XRP to 18.7% of the basket (target 9.5) and HBAR to 14.8% (target 5.25), and the quarterly discipline cut both back in one 21.2% one-way event — the largest in the index's window — while restoring BTC from 15.7% to 20% and ETH from 12.5% to 18%. Five quarters later, the April 1, 2026 event shows the same rulebook in a quiet tape: 3.7% one-way, the largest single change trimming TRX from 7.8% back to its 5.25% target. Same rule, an order of magnitude apart in traded size — the event record is what makes that variance visible, because the 8.1% average conceals it.

OCOPS3, the outlier. Crypto Ops 3 is a three-name basket of crypto-operations equities (targets 40/30/30), quarterly, 414-day window from 2025-06-05 — and the book's turnover outlier at 17.3% average one-way per event, 76.4% annualised. The reconstruction shows why: with three names this concentrated and this dispersed, single constituents run far past target between quarterly resets. At the July 1, 2025 event, BMNRon had reached 64.1% of the basket and was cut back to 30 in a 34.1% one-way event; at October 1, 2025, IRENon had reached 63.9% and was cut to 40 (24.0% one-way). Implied drag is 0.153% a year at 10 basis points and 0.305% at 20 — the book's highest, and the honest price of quarterly discipline on a three-name basket. The Q3 review recommends the committee consider a wider rebalance band or a semi-annual cadence for this structure at the Q4 review; that is a recommendation, pending committee review.

The drift cohort's zero. The four drift-triggered indices — OCI6, ODR4, OBAL6, OAILEAD8 — declare a 25-percentage-point trigger and produced no events in any reconstructed window. Maximum drift since window start: 10.1 points (OCI6, 730 days), 8.6 (OBAL6), 7.8 (OAILEAD8), 1.8 (ODR4, 106 days). Three things make this zero significant rather than vacuous. First, it is an affirmative disclosure only a published reconstruction can make: the honest turnover figure for these products over this window is zero, with zero implied cost drag, and the ledger shows the rule was evaluated daily and never fired — absence of trades as data, not as missing data. Second, it calibrates the threshold publicly: over a window that forced a 21% one-way event on quarterly Core 10, the 25-point band absorbed every excursion at under half its trigger — what a band designed to trade only on genuine dislocation should show. Third, it is consistent with the published methodology, which reported the same overlay not firing across the two-year Diversified window — the ledger turns that one-line remark into an inspectable event history (or the inspectable lack of one).

4. The honesty framework

The reconstruction disclosure. The ledger's central honesty obligation is that none of this happened. The events are what the stated rulebook implies over historical prices, computed by simulation; there were no orders, no fills, no venue, no counterparty. Every user-facing rebalancing surface carries the disclosure verbatim:

"Rebalance history is reconstructed by simulating the index rulebook over historical constituent prices; no live trades were executed. Past performance is not a guarantee of future results."

The disclosure is enforced at the vocabulary level, not just the footer level. The program maintains a use/avoid register for every rebalancing surface:

Basis alignment, disclosed per index. Each ledger records both the declared policy and the cadence under which the index's published statistics were actually simulated, and a boolean flag — basis-aligned — that is false whenever the two differ. As of this generation, 14 of the 28 tradables are misaligned: their stated policy is monthly or drift-triggered, while the analysis pipeline that produces the published statistics simulates quarterly. The product surfaces this with a templated note naming both cadences and stating that figures will be restated when the simulation is aligned with the stated policy. This flag is the ledger's most important honesty feature: a surface that reconstructed the declared policy while sitting silently beside statistics simulated under a different one would be more misleading than no surface at all. The misaligned count is a metric to drive to zero through pipeline alignment, and it is published while nonzero.

Limitations, stated plainly. The reconstruction inherits the backtest's assumptions and adds none of its own. Execution is frictionless: events fill in full at the observed UTC close of the boundary day, with no bid-ask spread, no market impact, no slippage, and no latency — the NAV series and all published statistics are gross of costs, and the cost figures the ledger does publish are a sensitivity overlay from a stated linear convention, not a fill-by-fill estimate; real execution on thin constituents would be worse. Missing bars forward-fill, and each index's window is the intersection of its constituents' histories, so annualised turnover on young baskets is noisy — the youngest monthly ledger annualises four events over 130 days. Drift is measured on the simulation basis, not on any live holding. And the reconstruction carries the backtest's survivorship caveat: today's constituents are the reconstructed constituents.

Surface discipline. The same standard governs what does not get a ledger. Retired indices have no rebalancing surface; non-tradable catalogue entries have none; the one tradable whose declared cadence the engine cannot simulate is skipped with its reason published rather than approximated under a cadence it never declared. A reconstruction standard is only as credible as its refusals.

5. Forward path

The 2026-Q3 review proposes — and this paper reports as proposals, pending index-committee review — a rebalance calendar that would attach announcement discipline to the reconstruction discipline already in place.

Proposed calendar. Effective dates at calendar boundaries: the first of the month or quarter, 00:00 UTC. The rationale is the basis-alignment property of Section 4: UTC calendar boundaries are what the engine already simulates, so adopting them as stated policy makes published statistics and stated policy share one rulebook by construction, where any imported convention (a third-Friday effective date, for instance) would misalign every published figure until the engine was re-simulated. On top of the boundary dates, an announcement layer in the institutional pattern: each scheduled rebalance announced five calendar days ahead with pro-forma weights, targets frozen one day before — late data rolls to the next cycle rather than moving a frozen vector. For the drift-triggered cohort, a published daily check at 00:00 UTC against the 25-point threshold, with current drift levels published alongside the trigger. A migration of the equities-settled cohort to the underlying venue's third-Friday convention was considered and explicitly deferred: it would break basis alignment, and the review says so rather than proposing it quietly.

The same review contains adoption and relabel proposals for the computed-weight indices, with a possible off-cycle corrective attached; those remain conditional on pre-registered gate runs and committee review, and this paper deliberately announces none of them.

Regeneration discipline. A published reconstruction that lags its catalogue is a new form of the old dishonesty, so the ledger is chained to the pipeline: after any adoption or weight change, the analysis pipeline re-simulates, the exports regenerate, and the ledger regenerates with them — in that order, per the review's checklist — so the event history never describes a rulebook the catalogue no longer states. The standing metrics for the program are three: the basis-misaligned count (14 of 28, to be driven to zero through pipeline alignment), the turnover outlier watch (one index above 65% a year, under review), and the coverage of the tradable book (28 of 29, pending an engine strategy for the session cadence).

The desk's summary position: a simulated index product cannot publish an execution record, but it can publish the full consequences of its stated rules, under its stated conventions, with its misalignments named — and it should be judged on whether it does.

Status: published reconstruction; forward calendar and all governance items proposed, pending index-committee review. All ledger figures follow the July 23, 2026 generation (2-year crypto history basis, generated 22:13 UTC, after that day's final availability regeneration); adjacent desk documents citing the same day's earlier run may differ in the second digit. Cost figures follow the published convention — two times one-way annualised turnover times the execution assumption, both legs, no market impact — at 5, 10, and 20 basis points. No live trades were executed; every figure in this paper is a backtest reconstruction.