Systematic Alpha Research Lab  ·  Signal Validation ReportAs of 2026-08-14
Research mandate

Does this signal
survive reality?

Testing whether systematic alpha signals survive costs, out-of-sample validation, and market stress. Each of 5 signals is treated as a research case and carried through one validation sequence.

Validation sequence
  1. 01Thesis
  2. 02Cost-aware evidence
  3. 03Out-of-sample
  4. 04Walk-forward
  5. 05Robustness
  6. 06Classification
Case 01 / 05Time-Series Momentum
Survived

Five signals under review

Each row is a research case carried through the same validation sequence. Select one to open its file; the sections that follow update to the selected signal.

Signal & thesisClassificationKey evidencePrincipal weakness
01
Time-Series Momentum
Assets in a sustained year-long uptrend tend to continue; the strategy holds those with positive medium-term trend.
SurvivedSharpe 0.57 · DD -15%
CAGR 5.7% net
Performance is uneven across market regimes.Viewing
02
Cross-Sectional Momentum
The strongest-trending assets relative to the universe tend to keep leading; the strategy holds only the relative leaders.
SurvivedSharpe 0.62 · DD -18%
CAGR 8.8% net
Performance is uneven across market regimes.Open
03
Short-Term Reversal
Assets with a sharp multi-day sell-off tend to partially recover; the strategy holds the most oversold names.
RejectedSharpe 0.44 · DD -29%
CAGR 5.1% net
High turnover (12.8x/yr) makes it execution-sensitive.Open
04
Volatility-Scaled Momentum
Trend-following with position sizes scaled down as volatility rises, holding portfolio risk closer to a fixed target.
ConditionalSharpe 0.57 · DD -12%
CAGR 4.3% net
Performance is uneven across market regimes.Open
05
Equal-Weight Signal Ensemble
Combine the four signals' standardized scores and hold the assets the blend rates positively, reducing reliance on any single signal.
SurvivedSharpe 0.59 · DD -22%
CAGR 7.8% net
Performance is uneven across market regimes.Open

Time-Series Momentum

Assets in a sustained year-long uptrend tend to continue; the strategy holds those with positive medium-term trend.

Survived5.7% CAGR net0.57 Sharpe-15% max drawdown3.2× turnover/yr
Signal thesis

Assets with positive medium-term trends tend to keep performing, because capital moves slowly, investors underreact, and macro regimes persist.

Definition

12-month return, skipping the last month.

momentum_12_1[t] = adj_close[t-21] / adj_close[t-252] - 1
Lookback  252 trading days (skip most recent 21)Rebalance  monthly
Portfolio construction

Hold every asset with positive 12-1 momentum, equal-weighted; park the rest in cash.

Weighting  equal weightConstraints  long-only · no leverage · cash fallback
Conditions for success
  1. Works in persistent trend regimes
  2. Avoids prolonged drawdowns by stepping aside
  3. Rotates across asset classes
Failure conditions
  1. Lags sharp reversals
  2. Underperforms in choppy markets
  3. Misses V-shaped recoveries

Performance, gross and net of costs

Growth of an index of 100 over the backtest sample (from 2007-02-28), against a selectable benchmark. The strategy is shown gross and net of 5 bps transaction costs.

Time-Series Momentum vs SPY Buy & Hold
Did the signal add value after realistic execution costs?
Emphasis
Benchmark
Strategy, net (5 bps)Strategy, grossSPY Buy & Hold
121713314906498082007201020132016201920222025
FindingNet of 5 bps, the strategy compounded at 5.7% per year, behind SPY Buy & Hold by -5.6% per year, at Sharpe 0.57 against 0.57.
CAGR, net
5.7%
Sharpe
0.57
Max drawdown
-15%
Volatility
7.6%
Turnover / yr
3.2×
Cost drag / yr
0.17%
Drawdown profile
How deep were the peak-to-trough losses?
-13%-11%-8%-5%-2%1%2007201020132016201920222025
FindingWorst peak-to-trough loss of -15%, against -55% for SPY buy-and-hold over the same window.
Out-of-sample check
Did it hold on data the parameters never saw?
In-sample, 2006 to 20165.3%0.66
Out-of-sample · 2017→6.1%0.47
CAGRSharpe
FindingOut-of-sample Sharpe 0.47 vs in-sample 0.66; held up out-of-sample.

Response to moving assumptions

Each module re-runs the signal under a different perturbation (cost, parameter, rebalance frequency, market regime, and crisis window) to test whether the result holds when the assumptions move.

Cost sensitivity
Do transaction costs erode the edge as execution gets more expensive?
Cost per unit turnover
Net @ 5 bpsGross
881321762202643092007201020132016201920222025
Net CAGR @ 5 bps5.7%
Net Sharpe0.57
Cost drag / yr0.17%
Cost test, this scenario only
viable, unchanged vs baseline
Final classification: Survived, determined by the full evidence (turnover and benchmark-relative result), not cost level alone.
0.000.170.330.500.660.581 bps0.575 bps0.5410 bps0.4825 bps
ResultFrom 1 → 25 bps, net Sharpe moves 0.580.48 and cost drag rises to 0.84%/yr. It still clears costs at the stress level.
Parameter robustnessone setting or many?
0.000.200.400.600.800.706M0.619M0.6112M0.5712M-1M
Result 4 of 4 parameter settings remain economically useful; the result is not a one-setting artefact.
Rebalance frequencytiming-dependent?
0.000.160.320.480.640.54Wee0.57Mon0.53Qua
Result weekly turnover (7.4×) against monthly (3.2×); net Sharpe is stable across frequencies (0.54 → 0.57).
Market regimeswhere it helps / hurts
-0.74-0.140.461.061.661.39Risk-On-0.55Risk-Off0.61High Vol0.43Rate Shock-0.38Inflation0.91Normal
Result strongest in Risk-On (Sharpe 1.39), weakest in Risk-Off (Sharpe -0.55); behaviour is regime-dependent.
Crisis windowsbehaviour under stress
-16%0%17%33%49%-2%GFC '08-2%Euro '11-3%COVID '20-11%Rates '2242%High-Rate '23+
Result worst window Inflation / Rate Shock: -10.7% cumulative (+7.5% relative to SPY).
Survived

Time-Series Momentum holds up: net Sharpe 0.57, -15.3% max drawdown, and out-of-sample Sharpe 0.47. It survives realistic costs and parameter variation.

Classification ledger

The committee classification for each signal, grouped by verdict and backed by the same evidence inspected above. Select any row to reopen its file.

Of five signals examined, three survived validation as documented research overlays, one is conditional, and one is rejected on this evidence.

SignalPrimary evidencePrincipal weaknessResearch note
Survived3 signals
Useful net-of-cost performance, controlled drawdowns, and credible out-of-sample evidence that does not depend on a single parameter choice.
Time-Series Momentum
Net Sharpe 0.57, max drawdown -15%, 5.7% CAGR net of 5 bps.
Use as a long-only de-risking trend overlay that controls drawdowns, not as a standalone return-maximizing strategy.
Net Sharpe 0.57 at 5 bps with -15.3% max drawdown (SPY drew down -55.2%).Performance is uneven across market regimes.Time-Series Momentum holds up: net Sharpe 0.57, -15.3% max drawdown, and out-of-sample Sharpe 0.47. It survives realistic costs and parameter variation.
Cross-Sectional Momentum
Net Sharpe 0.62, max drawdown -18%, 8.8% CAGR net of 5 bps.
Use as a relative-strength allocation sleeve within a diversified book, not as proof of standalone alpha.
Net Sharpe 0.62 at 5 bps with -18.4% max drawdown (SPY drew down -55.2%).Performance is uneven across market regimes.Cross-Sectional Momentum holds up: net Sharpe 0.62, -18.4% max drawdown, and out-of-sample Sharpe 0.72. It survives realistic costs and parameter variation.
Equal-Weight Signal Ensemble
Net Sharpe 0.59, max drawdown -22%, 7.8% CAGR net of 5 bps.
Use as a diversified, lower-variance core blend, not as a single-signal bet or proof of alpha.
Net Sharpe 0.59 at 5 bps with -21.8% max drawdown (SPY drew down -55.2%).Performance is uneven across market regimes.Equal-Weight Signal Ensemble holds up: net Sharpe 0.59, -21.8% max drawdown, and out-of-sample Sharpe 0.59. It survives realistic costs and parameter variation.
Conditional1 signal
Adds value under specific regimes or cost levels, or as a risk-control overlay. Useful, but not standalone alpha.
Volatility-Scaled Momentum
Net Sharpe 0.57, max drawdown -12%, 4.3% CAGR net of 5 bps.
Use as a volatility-control overlay when drawdown control matters more than raw return, not as a return maximizer.
Net Sharpe 0.57 at 5 bps with -11.8% max drawdown (SPY drew down -55.2%).Performance is uneven across market regimes.Volatility-Scaled Momentum is conditional: it adds value mainly through risk control (net Sharpe 0.57, -11.8% max drawdown) rather than raw return, and depends on costs or regime.
Rejected1 signal
Costs, turnover, benchmark-relative weakness, or instability undo the signal. A documented negative result.
Short-Term Reversal
Net Sharpe 0.44, max drawdown -29%, 5.1% CAGR net of 5 bps.
Not recommended as a standalone allocation on this evidence; at most a low-cost tactical input, never a return-maximizing strategy.
Net Sharpe 0.44 at 5 bps with -29.1% max drawdown (SPY drew down -55.2%).High turnover (12.8x/yr) makes it execution-sensitive.Short-Term Reversal is rejected on this evidence: it does not clear simple benchmarks; net Sharpe 0.44, out-of-sample Sharpe 0.32.

Data, construction, and validation record

ETF prices from Yahoo Finance (dividend- and split-adjusted close). Macro series from Yahoo Finance (VIX, 10-year and 13-week Treasury yields) and FRED (CPI). NBER recession dates and fixed crisis windows are used only for retrospective regime labeling. Portfolio returns use adjusted close as a total-return approximation. Every figure on this page is produced by the Python research engine and read from generated JSON; none is entered by hand.

Data & universe
  • 15 liquid ETFs · 9 asset groups
  • Price panel 2006-01-032026-08-14
  • Backtest calendar from 2007-02-28
  • Returns: dividend- and split-adjusted close (used as a total-return approximation)
  • Sources: FRED, NBER, Yahoo Finance
No-lookahead & cash proxy

feature[t] uses data through t; signal[t] formed at t; position[t+1] uses signal[t]; return[t+1] earned by position[t+1].

Cash / T-bill exposure (the cash fallback and the cash benchmark) earns the 13-week US Treasury-bill yield from Yahoo (^IRX), applied continuously across the full sample. BIL and SHV are included as tradable ETFs, but the continuous T-bill rate is used for cash so there is no 2007 inception gap.

Cost model & turnover

Cost = turnover x bps / 10000, applied as a return drag.

Turnover = sum of absolute weight changes at a rebalance (total traded notional / NAV, including both buys and sells).

1 / 5 / 10 / 25 bps · primary 5 bps

Validation protocol
  • Train 2006-01 to 2016-12
  • Test 2017-01 to present
  • 4 expanding walk-forward windows
  • Cost · parameter · rebalance · regime · crisis

Regime and crisis labels are used for retrospective validation, not live trading decisions. Macro inputs are lagged so labels use only information observable at the time.