Quantitative Backtesting

EPS Revision Strategy

Factor research pairing analyst EPS-revision and price-momentum signals across a liquid basket of stocks — each month the universe is the top 100 US names by trailing average dollar volume (price × volume), rebalanced monthly with no look-ahead and ETFs/funds excluded. Backtest spans 2015–2024 (113–114 monthly rebalances), reported net of 5 bps one-way transaction cost.

Strategy comparison

Strategy Annualized Sharpe Max Drawdown Ending Index
Revision + momentum composite (top 20%) 18.6%1.01-30.7%499.4
Momentum only (6M, top 20%) 18.6%0.89-33.1%497.1
EPS revision only (top 20%) 13.9%0.78-31.5%345.6
QQQ (benchmark) 17.6%0.96-33.1%466.5
SPY (benchmark) 11.7%0.79-24.8%285.9

Index normalized to 100 at inception. Sharpe and drawdown computed on net monthly returns. The three top rows decompose the composite into its two standalone signals.

Key takeaway — momentum drives the return; EPS revision refines the risk. Decomposing the composite is the honest test: momentum alone already captures essentially all of the return (18.6%, identical to the composite), while EPS revision alone is weak — 13.9% at a 0.78 Sharpe, barely distinguishable from SPY. Revision is not a standalone alpha here. Its value is incremental: layered on momentum it lifts Sharpe from 0.89 to 1.01 and trims the drawdown from -33% to -31%, acting as a fundamental filter that drops momentum names analysts are not backing. The right framing isn't “revision is the edge” — it's “momentum earns the return, revision improves its quality.”

Worked example — NVDA

NVIDIA was in the liquid universe for 100 months. Each month the strategy asks two questions: are analysts raising next-year EPS for this name (top 20% of revisions cross-sectionally)? and is 6-month price momentum positive? When both are true, the stock enters that month's book. NVDA qualified in 37 of 94 evaluable months — and those marks cluster exactly along its run from ~$1.71 to ~$134 (split-adjusted). Momentum is what kept NVDA in the book through the trend; the revision test is the fundamental check that it wasn't momentum alone — analysts were raising estimates the whole way up.

$2$5$10$20$50$100 201620172018201920202021202220232024
NVDA price (split-adjusted, log scale) Month NVDA qualified for the book (top-20% revision & momentum > 0)

And here is the raw signal driving the revision test — NVDA's actual month-over-month change in consensus FY1 EPS. Analysts ratcheted estimates up almost every month, with sharp spikes at each AI-demand re-rating — +62.6% in May 2023 (the worked example below) and a +102% jump in Feb 2024 after the Q4 FY24 blowout. The downgrades cluster in 2018–19 and the 2022 drawdown — exactly when the qualifying dots above thin out.

+60%+40%+20%0%-20% 201620172018201920202021202220232024
Upward EPS revision (analysts raising FY1 estimate) Downward revision

Month-over-month % change in consensus mean FY1 EPS for NVDA, from the estimates history. 73 of 100 months were upgrades; bar height is the true magnitude (clipped y-axis labels shown to +60%, peak bar is +102%).

Zooming into one rebalance — 31 May 2023, right after NVIDIA's blowout AI-driven guidance — shows the mechanic:

EPS1 revision (1M)+62.6% — 99th percentile of the universe that month
6-month momentum+124% (price $16.9 → $37.8 since Nov 2022)
Both signals on?Yes → NVDA enters the June book
Next-month result+11.8% ($37.8 → $42.3), and it kept qualifying through Aug 2023

Cross-sectional percentile computed against all top-100 liquid names in the same month; momentum and forward return from split-adjusted month-end prices. Signals are known at the rebalance date, so selection uses no look-ahead.