Home / Projects / 01
CASE STUDY · EQUITIES

Repairing 15 years of ASX equities history

An institutional investment team found their back-test results drifting between analysts. The root cause was a historical price series that had never been properly adjusted for splits, dividends and delistings.

The problem

The client's systematic research process relied on a daily ASX price series built incrementally over fifteen years. As new analysts joined, each had spliced in a new source to fill gaps, but no one had gone back to repair the older records. The result: split adjustments were inconsistent across time, dividends were not uniformly reinvested, delisted names had been silently removed (introducing survivorship bias), and two analysts running the same back-test would get meaningfully different equity curves.

The team had tried to fix this internally twice, but each attempt was abandoned because the manual reconciliation work ran into weeks of effort. The dataset was also being used to support investment-committee risk numbers, so the inconsistency was not just an academic problem — it was affecting live position sizing.

What we did

We rebuilt the history from two independent source feeds, reconciling closing prices on a date-by-date basis. For each ASX-listed name we:

  • Applied a consistent split and capital-return adjustment convention across the full 15-year window.
  • Rebuilt dividend history from corporate-action announcements, not from adjusted-price back-outs.
  • Re-included delisted and merged names so the universe reflected what was actually investable at each point in time.
  • Flagged every date where the two sources disagreed by more than 0.5%, and documented the resolution rule.
  • Delivered a point-in-time version of the universe so that historical back-tests only saw names and prices knowable at that date.
15 yrs
ASX price history rebuilt across a single, consistent adjustment convention
8% → <1%
Spread between two independent back-test runs, annualised
2 feeds
Independent source series reconciled date-by-date, disagreements logged
PIT
Point-in-time universe so back-tests never suffer look-ahead bias

The outcome

The rebuilt dataset reduced the spread between two independent back-test runs from roughly 8% annualised to under 1%. The client's risk committee accepted the new series as the single source of truth, and the two internal analysts who had been maintaining parallel spreadsheets were able to retire them. We handed over a data dictionary, a reconciliation report and a known-issues register listing the small number of ASX names where source disagreement could not be fully resolved.

The engagement also produced a small ongoing maintenance job: each quarter we refresh the most recent two years and flag new corporate actions, so the team does not have to rebuild anything manually. The work that had stalled internally for two attempts was completed in a single focused engagement.

Note: This case study describes data cleansing only. The client's investment process remains their own, and we did not advise on, or run, any strategy.