Complete Overhaul: Signal-Guided Re-Ranking

The Complete Overhaul tool sits in the Autopsy mode (not the Post-Mortem mode) of FM103, but it is fundamentally a diagnostic exercise. It takes a single rebalance period from a saved backtest and asks: "if I had used signal-derived factor weights instead of my chosen ones, what would the portfolio have looked like, and how would it have performed?" The exercise isolates the contribution of factor-weight choice from stock-pool choice and from optimization choice.

What "complete" overhaul means

The exercise re-ranks the entire investable universe (S&P 500 or whatever index was configured) under signal-derived factor weights instead of user-set weights, selects a new top-N, applies the same optimizer the original backtest used (equal-weight for FM101, or MVO/HRP/IVOL for FM102), and compares the resulting period equity path against the original.

This is different from the Weight Analysis tool, which keeps the same holdings and only re-weights them. Complete Overhaul replaces both the holdings AND the weights.

Where the "signal-derived weights" come from

The signal engine combines four cross-sectional macro signals available at the period start date (no look-ahead):

• Regime signal: low-vol bull favours growth + momentum; high-vol favours quality.
• Valuation spread signal: if top-quintile valuation is unusually cheap vs. bottom, weight value up.
• Factor momentum signal: factors that worked in the recent prior periods get higher weight (factor momentum, see Babu, Levine, Ooi, Pedersen & Stamelos 2020).
• Dispersion signal: when cross-sectional dispersion is high, factor strategies have more to work with — weights are amplified.

The signals are combined into composite V/Q/M/G weights summing to 100%.

The five-step pipeline

1. Compute signals at the period start date from public macro data + prior-period holdings.
2. Convert signals to factor weights via the platform's composite formula.
3. Re-rank the universe using new weights through FM101 or FM102's rescore endpoint.
4. Select the top-N and apply the original optimization method.
5. Compute period returns on the new portfolio and compare against the original.

The comparison metrics

The output table reports for both original and overhauled portfolios:

• Period return
• Holdings count and composition (Jaccard overlap)
• Sector breakdown
• Daily equity path (overlaid chart)
• Added / dropped / retained ticker lists

Interpreting the result

Three patterns:

• Overhaul return > original by a wide margin and most tickers changed: The user's factor weights were materially suboptimal for this period; signal-derived weights would have substantially helped. This may suggest a regime mismatch.
• Overhaul return ≈ original and most tickers similar: User's weights were close to the signal-implied. Period would have looked similar.
• Overhaul return < original: User's weights captured something the signals missed. Possibly skill, possibly luck — check across many periods before concluding.

Use case: post-mortem on a bad period

The most useful application: a backtest period where the strategy underperformed. The overhaul asks "would different factor weights have helped?" If yes, the underperformance was a weight-allocation issue (potentially fixable with a regime-aware overlay). If no, the underperformance was a stock-pool issue (the optimizer + signals could not have selected better names from the available universe).

Caveats

• One-period exercise. Overhaul on a single period is informative but not a strategy validation. The signal-derived weights may have been worse in most periods even when they were better in this one.
• Signal engine is a model. The four-signal composite is platform-specific and may not match the user's mental model of factor selection.
• Universe re-ranking is computationally expensive. Each overhaul triggers a full universe rescore in FM101/FM102, taking 30–90 seconds.

What overhaul does not replace

Complete Overhaul tells you what would have happened with different weights in one period. It does not:

• Test whether signal-derived weights work across all periods (that's a separate, longer experiment).
• Validate the signal engine itself (the four signals are heuristics, not guarantees).
• Identify which signal mattered (it computes only the composite outcome).