Where the price history comes from, and where it is spliced
The Multi-Source Historical Price Pipeline
Cycle-tested research requires daily price history that spans multiple regime turns. A single-venue history sourced from Binance Spot reaches back to 2017 for the largest crypto majors and to 2018-2020 for everything else. That is one full cycle of usable data per asset, and zero usable data on the pre-Binance era when crypto first traded against fiat at meaningful liquidity. This paper documents the methodology behind OLTA's multi-source price cascade, the splice rule that arbitrates between venues at the listing boundary, and the multi-anchor convention OLTA uses to report Bitcoin's risk-adjusted profile.
May 25, 202610 min read·7 sections
Refreshing constituents that a cycle left behind
Correcting Construction Drift After a Cycle
Many crypto baskets in 2026 still carry constituents whose original thesis was written for the 2021-2022 cycle. The constituents drew down 80% or more from cycle highs, never recovered, and continue to dilute basket performance against the assets that did survive. This paper documents the institutional principle behind a post-cycle constituent refresh, the framework OLTA applied across the May 2026 catalogue refresh, and the trade-offs an allocator should understand when a basket is reconstituted rather than rebalanced.
May 25, 202610 min read·8 sections
Why a basket can only be tested as far back as its youngest leg
The Cycle-Tested Construction Doctrine
Most crypto baskets cannot be backtested through a full cycle because their youngest constituent dates back only to 2020 or later. The intersection-of-availability rule on which any honest backtest is built forces the basket window to start at the youngest constituent's listing date, and for most baskets that date is inside the most recent cycle. This paper documents the cycle-tested construction principle, the reference basket OLTA built to make the principle operational, the mathematical limits a cycle-tested basket faces against BTC, and the implications for the catalogue's overall composition.
May 25, 202610 min read·8 sections
When the hundredth of a percent matters, and when it does not
Weight Precision in Index Construction: When the Hundredth of a Percent Matters
An index publishes a target weight for every constituent, and the number of decimals it publishes is a design decision, not a formatting detail. The OLTA catalogue carries two kinds of weight: committee-set allocations expressed in whole percentage points, and computed targets produced at full floating-point precision by the hierarchical-risk-parity, equal-risk-contribution, and inverse-volatility pipelines. This paper quantifies what is lost when a computed target is rounded for publication. At a 0.1% grid, the rounding artefact contributes roughly 5 basis points a year of tracking noise against the optimiser's own vector — the same order of magnitude as the implementation-cost figures OLTA publishes in its rebalancing ledger. At a 0.01% grid the artefact falls to roughly half a basis point, an order of magnitude below anything on a published surface. We review the divisor mechanics that make the grid choice invisible at the NAV level, the institutional convention (S&P Dow Jones Indices, MSCI, and the major ETF issuers publish constituent weights to the hundredth of a percent), and the current display state of the catalogue. The desk recommends adopting the hundredth as the publication grid for computed targets and for display, under a four-clause rounding rule whose centrepiece is a signature convention: committee weights remain whole numbers, and a decimal on a published weight means the number came out of a computation. The recommendation is pending ratification by the index committee. Committee-weighted funds are unaffected in every scenario.
Jul 24, 202613 min read·5 sections
HRP, ERC and inverse-volatility as running code, not labels
From Labels to Computation: HRP, ERC and Inverse-Volatility Weights in Practice
Several indices in the OLTA catalogue carry weighting labels that name an algorithm — Hierarchical Risk Parity, Equal Risk Contribution, Inverse Volatility. A label of that kind is a methodological claim, and the desk's standard is simple: a weighting label must describe code that runs. This paper documents the computation stack now running behind those labels — a returns pipeline on a 180-day lookback, Ledoit-Wolf shrunk covariance, HRP by recursive bisection, ERC by fixed-point iteration, inverse-volatility weighting, and a constraint pass for cash sleeves, single-name caps and floors. It publishes the first five candidate target vectors next to the committee-set weights currently in production, reads the gaps between them, and sets out the pre-registered walk-forward discipline that will govern adoption. No adoption is announced here. The evaluation gate has not yet been executed, every production weight remains committee-set as of this date, and each candidacy is pending index-committee review. The outcome set is fixed in advance and contains exactly two exits: adopt the computed weights, or relabel the index. Keeping a computed-sounding label on weights no algorithm produced is not among them.
Jul 24, 202620 min read·6 sections
Every reconstructed rebalance, dated and costed
Reconstructed Rebalancing Ledgers: A Transparency Standard for Simulated Index Products
An index that executes trades leaves a rebalancing record as a byproduct: announcement files, pro-forma weights, effective dates, and the turnover the trades realised. A simulated index product leaves nothing, which is precisely why the record has to be manufactured deliberately or it will never exist. This paper documents the OLTA rebalancing ledger: for each of the 28 tradable indices in the catalogue, a reconstruction of every rebalance event the index's stated rulebook implies over its full backtest window — pre- and post-event weight vectors, one-way turnover, NAV at the event, and a cost-sensitivity table under a stated execution convention — produced by the same engine, divisor mechanics, and data preparation that generate the published performance statistics. Every figure is a backtest reconstruction, not an execution record: no live trades were executed, each ledger says so in a machine-readable provenance block, and Section 4 develops the disclosure. The July 23, 2026 generation reconstructs 533 events across 24 indices; the four drift-triggered indices show zero events, itself a published finding. Annualised one-way turnover spans 5.6% to 76.4% across the book, and the implied return drag at a 10-basis-point execution assumption spans 0.011% to 0.153% a year. Fourteen of the 28 ledgers disclose that their published statistics were simulated under a different cadence than the stated policy — a misalignment named per index rather than left for readers to discover. The paper closes with the forward calendar proposed at the 2026-Q3 review — calendar-boundary effective dates, advance announcement with pro-forma weights, a pre-effective freeze — all of it pending index-committee review; nothing here announces a governance decision that has not been taken.
Jul 24, 202614 min read·5 sections
Delistings, rebrands, and the stale-tail problem
Index Continuity Through Constituent Migrations: Delistings, Rebrands, and the Stale-Tail Problem
An index is a claim about a basket of instruments through time, and the instruments do not hold still. Tokens delist, redenominate, rebrand, and list on dates that have nothing to do with one another, and every one of these events attacks the same assumption a backtest quietly makes: that a ticker names one instrument with one continuous history. This paper documents three failure modes the OLTA catalogue met in production this quarter — the hard termination (MKR, dead on the venue September 15, 2025), the rebrand-continuation (TON, renamed GRAM at one-for-one terms in June 2026), and the listing gap (MKR and MORPHO, two constituents of the same baskets whose venue lifetimes never overlapped) — and the incident that motivated a structural fix: for three weeks, eight indices published "two-year" statistics that had silently stopped on June 30, because a delisted constituent's dead tail dragged the right edge of every containing basket's simulation window back to its last trade, with no error raised and no disclosure on any surface. The remedy is two mechanisms with deliberately narrow mandates. An availability rule declares any series whose last bar sits more than five days behind the dataset's right edge unavailable: dropped from simulation, weights rescaled, coverage disclosed — and, asymmetrically, fatal to weight computation, because a computed target on a dead asset describes a basket that cannot exist. A continuation splice joins a legacy ticker to its verified one-for-one successor under a boundary guard that refuses any splice whose cross-gap price ratio leaves [0.8, 1.2]; the TON-to-GRAM boundary verified at 1.0000. The paper closes on the governance boundary the mechanisms must not cross: continuity of data is not continuity of product, and a data alias must never silently rename what the committee publishes.
Jul 24, 202616 min read·5 sections
