Reconstruction
Historical snapshots were aligned into a point-in-time ranking calendar with separate factor, score, rank, and return-observation layers.
Case Study
Reconstructing a Global Equity Ranking Model
A proprietary historical ranking system was rebuilt from point-in-time snapshots, cleaned for vendor-data issues, and evaluated with global, country-neutral, and sector-neutral forward-return diagnostics.
Research Problem
The client problem was not simply to run a backtest. The task was to determine whether an old ranking process could be reconstructed faithfully, whether the data pipeline was clean enough to trust, and whether the resulting rankings separated future winners from losers.
Cleaning
Source rows were checked for duplicate tickers, invalid markers, mixed decimal formats, missing values, and target-window drift.
Evaluation
Backtests compared top-ranked and bottom-ranked groups globally and inside country and sector neutral groupings.
What Was Tested
Model Inputs
The ranking model combined normalized factor families covering valuation, profitability, growth, balance-sheet strength, shareholder yield, size, and momentum. Exact coefficients and field mappings are proprietary and not disclosed.
Backtest Design
Rankings were evaluated using later return snapshots, capped target-window lag rules, and winsorized return summaries to reduce the impact of extreme single-name outcomes.
Data Integrity Controls
Point-in-Time Rows
Clean selected rows were built from historical vendor tables, excluding non-security header rows and resolving duplicate ticker/date rows.
Numeric Formats
Decimal-comma values were parsed as decimals rather than thousands separators, avoiding distorted factor values and return observations.
Return Windows
Forward return targets were capped by acceptable lag, preventing a nominal one-month test from drifting into a much later observation window.
Representative Results
The table shows winsorized top-minus-bottom spreads for the primary baseline reconstruction across multiple model-year buckets. Results are percentage-point differences between the top-ranked and bottom-ranked groups; country- and sector-neutral quintiles use top 20% versus bottom 20% within each group.
| Test | Model Bucket | Horizon | Top Group | Bottom Group | Spread | Obs. Top / Bottom |
|---|---|---|---|---|---|---|
| Global decile | 2020 | 1Y | 29.40% | 3.92% | +25.48 pp | 1,552 / 1,559 |
| Global decile | 2020 | 6M | 9.32% | -0.08% | +9.40 pp | 2,165 / 2,176 |
| Global decile | 2019 | 1Y | 22.99% | 15.23% | +7.76 pp | 1,537 / 1,544 |
| Global decile | 2019 | 6M | 6.15% | 4.29% | +1.86 pp | 2,148 / 2,157 |
| Global decile | 2018 | 1Y | 19.79% | 15.19% | +4.60 pp | 1,536 / 1,544 |
| Global decile | 2018 | 6M | 5.74% | 4.81% | +0.94 pp | 2,144 / 2,154 |
| Global decile | 2017 | 1Y | 23.72% | 14.85% | +8.87 pp | 1,533 / 1,541 |
| Global decile | 2017 | 6M | 6.43% | 4.47% | +1.96 pp | 2,140 / 2,149 |
| Sector-neutral quintile | 2020 | 1Y | 28.29% | 5.53% | +22.76 pp | 3,066 / 3,148 |
| Sector-neutral quintile | 2020 | 6M | 8.04% | 0.46% | +7.58 pp | 4,285 / 4,388 |
| Sector-neutral quintile | 2019 | 1Y | 21.40% | 13.84% | +7.56 pp | 3,037 / 3,115 |
| Sector-neutral quintile | 2019 | 6M | 5.44% | 3.80% | +1.63 pp | 4,248 / 4,345 |
| Sector-neutral quintile | 2018 | 1Y | 20.31% | 14.73% | +5.58 pp | 3,033 / 3,110 |
| Sector-neutral quintile | 2018 | 6M | 5.43% | 4.33% | +1.10 pp | 4,239 / 4,340 |
| Sector-neutral quintile | 2017 | 1Y | 22.49% | 14.63% | +7.86 pp | 3,025 / 3,105 |
| Sector-neutral quintile | 2017 | 6M | 6.32% | 3.70% | +2.62 pp | 4,232 / 4,324 |
| Country-neutral quintile | 2020 | 1Y | 26.58% | 6.68% | +19.90 pp | 2,923 / 3,125 |
| Country-neutral quintile | 2020 | 6M | 8.05% | 0.72% | +7.33 pp | 4,076 / 4,365 |
| Country-neutral quintile | 2019 | 1Y | 20.23% | 15.49% | +4.74 pp | 2,893 / 3,082 |
| Country-neutral quintile | 2019 | 6M | 5.57% | 4.36% | +1.21 pp | 4,043 / 4,310 |
| Country-neutral quintile | 2018 | 1Y | 18.33% | 16.48% | +1.85 pp | 2,889 / 3,081 |
| Country-neutral quintile | 2018 | 6M | 4.91% | 5.48% | -0.57 pp | 4,041 / 4,310 |
| Country-neutral quintile | 2017 | 1Y | 21.82% | 15.06% | +6.76 pp | 2,886 / 3,076 |
| Country-neutral quintile | 2017 | 6M | 6.40% | 3.60% | +2.79 pp | 4,032 / 4,295 |
Historical backtest results are shown for research discussion only. They exclude transaction costs, taxes, liquidity constraints, implementation constraints, and other real-world frictions.
Interpretation
Medium-Term Separation
The strongest separation appeared at six-month and one-year horizons, while one-month results were weaker and noisier.
Neutralization Matters
Sector- and country-neutral tests help distinguish stock-selection signal from broad geographic or sector exposure.
Variants Are Testable
Controlled model variants can be compared without disclosing proprietary rules, helping separate intuition from measurable effect.