Backtest Report · 1996–2024

Twenty-nine years. One hundred and eighty-one picks. One losing year.

The full performance record of the Mīzān selection rule, applied mechanically to the US universe every April for nearly three decades, at its primary five-to-seven-name concentration. Every figure on this page is computed directly from the research dataset. Structural limitations are disclosed in full below — read them carefully before interpreting any number.

29Performance years 181Individual picks 5–7Names per year Apr 2026Last verified
Before you read the numbers: the results presented here are extraordinary, and extraordinary results deserve extraordinary scrutiny. A full, honest accounting of the backtest's structural limitations — survivorship bias, look-ahead on forward estimates, retrospective regime classification, absence of transaction costs and taxes — is in the caveats section at the bottom of this page. Sophisticated readers should read it first.

How a stock makes the rule.

A high-level walk-through of the filter architecture. The order of operations and the philosophy behind each layer are disclosed below. The exact thresholds, weights, and regime triggers are proprietary — the standard disclosure boundary for a private quantitative system.

01 · Universe
US-listed, alive April 1
Liquid enough to deploy real capital. Micro-caps below a tradeable threshold are excluded. The universe rebuilds each April.
02 · Regime
Recovery or Normal
The prior calendar year decides the mode. Drawdown years route into Recovery — earnings-rebound logic. Calmer years route into Normal — value-with-quality logic.
03 · Quality gate
Solvency before opportunity
Names that fail the safety filter never enter the rank, regardless of how attractive the rest of the profile looks. The filter exists to eliminate the catastrophic single-name outcome.
04 · Rank
Composite, regime-specific
Surviving names are scored on a combination of operating quality, valuation, and momentum. The exact weighting differs between Recovery and Normal modes, and is the proprietary core of the rule.
05 · Cut
Top 5–7 names · equal weight
The portfolio takes the highest-ranked names from the composite. The cut-off is rules-based, not discretionary. Each name receives an equal allocation.
06 · Hold
Twelve months · no trading
No additions, no trims, no stop-losses. The portfolio is rebuilt only on the next April 1. The discipline is the design.
Why the formulas are not on this page. The full backtest record — every year, every pick, every return — is auditable below. The selection mechanics — thresholds, weights, regime triggers — are not. This page asks a reader to evaluate the rule’s behaviour over time, not to reverse-engineer the rule itself.

Backtest summary.

Concentrated 5–7 name configuration, 1996–2024. The aggregate behaviour of the rule across 29 holding years and 181 individual picks. Risk-adjusted decomposition follows in § 04; intra-year path in § 03; capacity treatment in § 09.

Compound Annual
70.30%
vs 10.11% for the S&P 500 over the same window. Annual compounding across 29 holding years; the daily-reconstructed path (70.20%) is reported in § 03. Theoretical — before slippage, taxes, and capacity drag.
Positive years
28 / 29
Single negative year: 2008 at −0.57%. Path is not as smooth as year-end totals suggest — intra-year drawdowns are now reported in § 03.
Beat the market
27 / 29
93% of holding years closed ahead of the S&P 500 total return over the identical window.
Worst Drawdown
−52.77%
Peak 2020-02-20 → trough 2020-03-18 (27 days). Full recovery 61 days later, on 2020-05-18. Total drawdown episode: 88 calendar days. Daily-derived; not visible in annual frequency reporting.
Daily Sharpe
1.72
Annualised from daily log returns, rf = 3%. Higher than the 0.91 annual-frequency Sharpe because the year-on-year stddev is dominated by 2-3 outlier vintages (2020, 2023); daily smooths across roughly 252 days.
Daily Sortino
2.52
Downside-only risk measure on the daily series, rf = 3%. The asymmetric upside-skew of the strategy makes Sortino materially higher than Sharpe.
Theoretical compounding
$10,000 → $50.7B
Mechanical compounding of the 29 annual returns at the realised 70.30% CAGR. No slippage, no taxes, no capacity constraint. This figure is what the rule would have produced if a single dollar could be deployed at every April 1 close at any size — which it cannot.
Capacity-bound reality
~$300M EW · up to ~$57B aggregate
Two liquidity ceilings implied by the rule. Equal-weight: $145M–$700M depending on configuration (Concentrated·7 at moderate participation = $300M; constrained by the marginal pick’s ADV at strict equal-weighting). Cohort-aggregate at ADV-proportional sizing: $5B (Concentrated) → $44B (Broad-20) → $57B (Broad-40) at moderate participation. The first reading is the rule applied at strict equal weight; the second is the rule applied with positions sized to per-name liquidity. Full bands across both ceilings, all three configurations, and three participation regimes in § 09.

Annual returns, daily path, intra-year drawdown.

The same 29-year window viewed at three time-scales. Annual bars frame the year-on-year shape. The daily equity curve is a 7,301-day reconstruction with shares fixed at each April 1 and drift between rebalances — the same accounting the live portfolio uses. The drawdown panel below the curve is the path metric annual frequency reporting cannot show.

Mīzān vs S&P 500 — Annual Returns (%)
MīzānS&P 500

The S&P 500 is the global default benchmark for US-listed equity strategies; total return is used throughout (dividends reinvested) and the comparison window is identical to Mīzān's holding period — 1 April to 1 April of the following year.

Cumulative Growth of $10,000 — annual close
MīzānS&P 500

Logarithmic scale. $10,000 hypothetically invested at the start of 1996, compounded according to each strategy's annual returns. The final Mīzān figure is theoretical — capacity constraints at scale are discussed below.

Daily equity curve — three configurations · S&P 500 overlay
Concentrated 5-7Broad-20Broad-40S&P 500

Logarithmic scale. NAV samples at month-end across the full 1996-04 – 2025-03 window for each of the three rule configurations, plus the S&P 500 (SPY) on the same $10,000 base. Daily NAV is reconstructed from each year's picks held with shares fixed (equal weight at entry, drift between April rebalances) — the identical mechanic the live Concentrated portfolio uses. CAGR over the full 7,301-day series: 70.20% Concentrated, 51.22% Broad-20, 41.37% Broad-40, 9.46% S&P 500.

Benchmark construction. Pre-1999-11 the S&P 500 line uses monthly price-only index closes from a contributed historical series, anchored to the same $10,000 base on 1996-04-01; from 1999-11-01 onwards it switches to SPY adjusted close (total return) at daily resolution, scaled at the bridge so the line is continuous. The pre-1999-11 cumulative therefore understates true total return by roughly the period's average dividend yield (≈2% per annum). All SPY-relative ratios reported in § 04 (β, α, IR, TE, captures) use only the 1999-11-01 → 2025-03-31 daily overlap (6,391 trading days), which is the window where both the strategy NAV reconstruction and SPY adjusted close are simultaneously available.

Underwater drawdown — % from running peak · S&P 500 overlay
Concentrated 5-7Broad-20Broad-40S&P 500

Drawdown at any point in time is the percentage decline from the running peak NAV. Read this as the answer to the question: “If I had bought into the strategy at its high water mark on day X, what is the worst paper loss I would have lived through before recovery?” Peak-to-trough worst on the daily series: −52.77% Concentrated, −50.06% Broad-20, −45.06% Broad-40 — all attained 2020-03-18 in the COVID shock. The S&P 500 line shows its own deepest drawdowns on the same axis: −50.8% in the 2007–2009 GFC and −33.7% in the 2020 COVID episode. Long curve uses month-end sampling with the February-to-May 2020 window resolved at daily granularity so the worst drawdown shows at its true depth.

February–June 2020 — the COVID shock at daily resolution
ConcentratedBroad-20Broad-40

The largest drawdown episode on file, plotted at daily resolution and indexed to peak NAV = 100. Twenty-seven calendar days from peak to trough (2020-02-20 → 2020-03-18); sixty-one to recovery on the Concentrated configuration. Annual reporting shows perf-year 2020 closed at +394.5% — an entirely different number from what the path of the year felt like.

Crash recovery — Mīzān vs S&P 500 · the four canonical bear markets

Read across each row. The peak-to-recovery window on the left is the S&P 500's full underwater stretch — the time an SPY investor entering at the prior high had to wait to be made whole. The right side reports what Mīzān was doing over that same window: the deepest intra-window drawdown and the cumulative return from SPY's peak date to SPY's recovery date. Drawdown depth on the strategy side is real; what differs from the index is duration. SPY required 6.6 years to recover from the dot-com bear and 4.9 years from the GFC. Mīzān's longest single underwater stretch over the same 1999-11 → 2025-03 window is 304 days.

Episode S&P 500 Mīzān Concentrated
Max DD Days underwater Peak → trough → recovery Max DD in window Return over SPY window
Dot-com bear
2000-03-24 → 2006-10-25
−47.5% 2,406 d (6.6 yr) 929 d → 1,477 d −22.5% +3,954%
Global Financial Crisis
2007-10-09 → 2012-08-15
−55.2% 1,772 d (4.9 yr) 517 d → 1,255 d −33.1% +683%
COVID shock
2020-02-19 → 2020-08-07
−33.7% 170 d (5.6 mo) 33 d → 137 d −52.8% +46.4%
2022 Inflation bear
2022-01-03 → 2023-12-12
−24.5% 708 d (1.9 yr) 282 d → 426 d −18.7% +225%

How to read. “Days underwater” is calendar days from SPY's prior all-time-high to its first new all-time-high; “Peak → trough → recovery” splits that into descent and ascent halves. “Max DD in window” is the deepest peak-to-trough Mīzān suffered inside the SPY window, measured from a running peak that resets at window start. “Return over SPY window” is the cumulative Mīzān Concentrated NAV return from SPY's peak date to SPY's recovery date — what compounded while the index was rebuilding.

The COVID asymmetry. COVID is the one episode where Mīzān went deeper than SPY (−52.8% vs −33.7%); it is also the episode where Mīzān recovered first. Mīzān total episode 88 days against SPY's 170; Mīzān back to all-time-highs 2020-05-15 vs SPY 2020-08-07.

Cross-reference. Aggregate path-shape metrics (% days at peak, average drawdown depth, longest single stretch, Calmar) for the full 1999-11 → 2025-03 overlap are reported in the Drawdown profile tier of § 04.

Return distribution.

Central tendency, extremes, dispersion, and risk-adjusted measures across 29 holding-period observations. Sortino, Information Ratio, alpha, and regression statistics computed at runtime from the published series.

Distribution
Mean Annual
+83.99%+11.73%
Arithmetic average of 29 holding-period returns.
Median Annual
+59.71%+15.80%
The typical year, half above and half below.
Best Year
+394.5%+33.4%
2020 vintage, Recovery mode.
Worst Year
−0.57%−37.0%
2008 — marginally negative; the only negative year on file.
Annual Volatility
89.5%18.0%
High by design. Asymmetric — almost all dispersion is to the upside.
Sharpe Ratio
0.910.48
Excess return per unit of volatility; rf = 3%.
Daily-path metrics · new in April 2026

A 7,301-day NAV reconstruction unlocks measurements that annual reporting cannot produce: peak-to-trough drawdown, time-to-recovery, and risk-adjusted ratios on the daily series. Equal-weight at entry, drift between April rebalances — the same accounting the live portfolio uses.

Daily Sharpe
1.720.23
Annualised from daily log returns, rf=3%. Higher than the 0.91 annual-frequency Sharpe because year-on-year stddev is dominated by 2-3 outlier vintages; the daily series smooths across roughly 252 trading days per year. SPY: 0.23 over the 1999-11 → 2025-03 daily overlap.
Daily Sortino
2.520.28
Downside-only risk on the daily series, rf=3%. The asymmetric upside-skew of the strategy makes Sortino materially higher than Sharpe. SPY's Sortino sits close to its Sharpe — symmetric distribution.
Daily vol (annlsd)
32.4%19.3%
Annualised stddev of daily log returns. Much lower than the 89.5% year-on-year stddev because the daily distribution is closer to normal once the outlier years are decomposed. SPY's daily vol over the same window: 19.3% — the strategy carries roughly 1.7× the index's daily noise.
Maximum Drawdown
−52.77%−55.20%
Worst peak-to-trough loss on the full 7,301-day series. Mīzān: 2020-02-20 peak → 2020-03-18 trough (88-day total episode). SPY: 2007-10-09 peak → 2009-03-09 trough during the GFC. Comparable depth — vastly different duration (next two cells).
Days to trough
27517
Calendar days from peak to bottom in the worst drawdown episode. Mīzān: COVID, 27 days. SPY: GFC, 517 days — the index took nearly 17 months to find the floor.
Days to recovery
611,255
From trough back to a new high. Mīzān recovered from COVID in 61 days (88-day total episode). SPY took 1,255 days from the GFC trough to a new all-time high — a 1,772-day total episode against Mīzān's 88. The depth is similar; the duration is the asymmetry.
S&P 500 overlap-period metrics · 1999-11-01 → 2025-03-31 · 6,391 trading days

Daily-frequency SPY-relative measurements computed across the 25.4-year overlap window where both the strategy NAV and SPY adjusted close are available at daily resolution. SPY data ingested from Alpha Vantage TIME_SERIES_DAILY_ADJUSTED. Strategy returns computed from the daily reconstructed NAV; SPY returns from adjusted close (total return). All ratios annualised from daily log returns, rf = 3%. Three columns — Concentrated 5-7 · Broad-20 · Broad-40.

Beta vs S&P 500
1.1491.052 · 1.009
Slope of strategy daily excess returns regressed on SPY daily excess returns. Concentrated has slightly more market sensitivity; Broad-40 sits effectively at market beta. Co-movement direction is real; magnitude is what beta measures.
Alpha · CAPM, annlsd
+72.65%+47.30% · +35.55%
Single-factor regression intercept on daily excess returns, annualised. The portion of return that is not explained by S&P 500 exposure at the strategy's beta. By construction this is where the rule's edge resides.
Information Ratio
2.112.32 · 2.24
Active return divided by tracking error. Reading: the strategy generates roughly 2.1–2.3 units of excess return for every unit of dispersion away from the index. IRs above 1.0 are considered exceptional in active-management literature.
Tracking Error · annlsd
24.46%15.60% · 12.57%
Annualised stddev of the strategy-minus-SPY daily return spread. Reads as “how unlike the index does the strategy behave?”. Concentrated runs the loosest tracking; the wider configurations sit closer to the benchmark.
Up-market capture
150.81%130.02% · 122.41%
Average daily return on days when SPY was up, divided by SPY's average up-day return. Above 100% means the strategy outperforms in rising markets. Concentrated captures roughly 1.5× the S&P's up-day move.
Down-market capture
102.75%95.80% · 95.67%
Same calculation across down days. Below 100% means the strategy fell less than SPY on average down days. Broad-20/40 deliver the more attractive asymmetry; Concentrated trades that down-side cushion for higher up-capture.
Correlation
0.6750.794 · 0.840
Pearson correlation of daily returns vs SPY. Concentrated holds 5–7 names so its day-to-day path diverges substantially from the index. The wider configurations correlate more closely as the position count rises.
S&P 500 reference
9.46% CAGR
SPY benchmark over the full 29-year strategy window 1996-04-01 → 2025-03-31. $10,000 compounded to roughly $137,562 — the apples-to-apples reference for the 70.20% Concentrated CAGR. Pre-1999-11 segment is S&P 500 price-only (no dividends reinvested) and post-1999-11 is SPY adjusted close (total return); the price-only stub understates true 29-yr SPY total return by roughly 0.3 percentage points/year. The daily-resolution metrics on the row above (β, α, IR, TE, captures, correlation) necessarily use the 25.4-year daily overlap (1999-11-01 → 2025-03-31, 6,391 trading days) because pre-1999-11 SPY adjusted close is only available at monthly resolution; SPY's CAGR over that shorter daily window was 7.65% — visible in the Calmar row below.
Risk-adjusted decomposition
Sortino (MAR=rf)
Mīzān figure flagged. Excess return per unit of downside deviation, measured against rf=3%. The Mīzān ratio is divided by the shortfall mass of a single annual observation in 29 years (the 2008 vintage at −0.57%, which sits 3.57 pp below MAR), so the divisor is statistically degenerate and the ratio inflates accordingly. SPY's denominator absorbs eight shortfall years, which is why its reading is finite and meaningful. Read the SPY figure as the anchor; treat the Mīzān number as a tail-shape signal (no losses against MAR), not a tradeable ratio.
Information Ratio
Active return over tracking error vs the S&P 500 across the 29 annual observations.
CAPM Alpha (annual)
Single-factor regression intercept on annual excess returns; rf = 3%. Beta and R² below.
Regression Beta
Slope of Mīzān excess returns against S&P excess returns. Direction with the market is real; magnitude of co-movement is what beta reports.
Regression R²
Fraction of Mīzān’s annual variance explained by the S&P. The unexplained majority is, by construction, where the rule’s edge lives.
Skewness
Third moment of the annual return distribution. Mīzān: positive (right-skewed) — the few extreme prints (2020 and 2023 Recovery vintages) sit on the upside. SPY: negative (left-skewed) — the few extreme prints sit on the downside (2002, 2008). The asymmetry direction is the structural difference, not the magnitudes.
Excess kurtosis
Fourth-moment tail weight relative to a Gaussian (zero = normal). Mīzān: heavily fat-tailed by design — the rule manufactures occasional triple-digit vintages alongside otherwise compact returns. SPY: near-normal at annual frequency — the daily series is famously fat-tailed, but annual aggregation washes most of it out.
Drawdown profile · path-dependent risk · 1999-11-01 → 2025-03-31

Aggregate path-shape across the 6,391-day daily overlap window. These ratios capture the experience of holding the strategy through time — not just point-to-point CAGR. Three columns — Concentrated 5-7 · Broad-20 · Broad-40 — with the SPY benchmark reading reported in each cell description. Max DD identical to the values shown in the daily-frequency reconstruction tier above.

% days at running peak
25.7%30.6% · 31.4%
Share of trading days within 0.5% of the running all-time-high. SPY: 18.5%. The strategy spends materially more time at fresh highs than the index — the wider configurations roughly 1.7× as much.
Avg drawdown depth
−5.1%−3.9% · −3.8%
Mean of daily drawdown values across all 6,391 days (zero on peak days). Equivalent to a path-weighted “pain index”. SPY: −11.1%. Roughly half the index's pain, across all three configurations.
Avg depth when underwater
−6.9%−5.5% · −5.5%
Mean drawdown depth on days the series is underwater (DD > 0.5%). Conditional on being in a drawdown, Mīzān's typical depth is half of SPY's. SPY: −13.6%.
Days underwater
74.3%69.4% · 68.6%
Share of days below the running all-time-high (more than 0.5%). SPY: 81.5% — the index is underwater four days in five over this window. Mīzān is closer to seven in ten.
Longest single stretch
304 d358 d · 362 d
Calendar days of the longest contiguous peak-to-recovery episode (deepest at >5%). SPY: 2,406 d — the dot-com bear, 2000-03-24 → 2006-10-25. The duration gap is the most consequential difference on this row.
Calmar ratio
1.401.04 · 0.92
Windowed compounding rate over absolute max drawdown for the 25.4-year daily overlap window (73.96% / 52.29% / 41.37% windowed rate over −52.8% / −50.1% / −45.1%). The 73.96% is the rate over this shorter daily-overlap window only — not the headline 70.30% full-record CAGR. SPY: 0.14 (7.65% / 55.20%, both measured on the same daily series). The full 29-year SPY CAGR is 9.46%; this row uses the shorter 25.4-year window so all three numerators come from a single daily NAV series. On this measure Concentrated runs roughly ten times the index's risk-adjusted compounding rate over the same window.
Year-by-year vintage results · Concentrated 5-7

Every Apr-1 entry → next Apr-1 exit, all 29 vintages on file. The portfolio is equal-weighted at entry and drifts between rebalances; vintage start dates use the first trading day on or after April 1. Mīzān Concentrated outperformed SPY in 27 of 29 years — the two underperforming vintages were 1998 (+18.4% vs SPY +28.6%) and 2021 (+25.7% vs SPY +28.7%). The single negative year was 2008 (−0.57%); every Recovery-mode vintage closed positive (worst +26.5%, see § 05). Mode classification is rule-based and proprietary; the behavioural profile of each mode is disclosed in § 05.

Vintage Picks Mīzān return SPY return Excess
Apr-96 → Mar-975+30.71%+23.00%+7.7
Apr-97 → Mar-985+100.08%+33.40%+66.7
Apr-98 → Mar-997+18.36%+28.60%−10.2
Apr-99 → Mar-007+40.92%+21.00%+19.9
Apr-00 → Mar-015+97.42%−9.10%+106.5
Apr-01 → Mar-025+97.66%−11.90%+109.6
Apr-02 → Mar-035+26.48%−22.10%+48.6
Apr-03 → Mar-045+101.76%+28.70%+73.1
Apr-04 → Mar-057+71.52%+10.90%+60.6
Apr-05 → Mar-067+124.13%+4.90%+119.2
Apr-06 → Mar-077+43.95%+15.80%+28.2
Apr-07 → Mar-087+110.72%+5.50%+105.2
Apr-08 → Mar-097−0.57%−37.00%+36.4
Apr-09 → Mar-105+131.23%+26.50%+104.7
Apr-10 → Mar-117+139.93%+15.10%+124.8
Apr-11 → Mar-127+10.03%+2.10%+7.9
Apr-12 → Mar-137+24.55%+16.00%+8.6
Apr-13 → Mar-147+59.19%+32.40%+26.8
Apr-14 → Mar-157+60.22%+13.70%+46.5
Apr-15 → Mar-167+10.63%+1.40%+9.2
Apr-16 → Mar-175+63.74%+12.00%+51.7
Apr-17 → Mar-187+47.62%+21.80%+25.8
Apr-18 → Mar-197+77.30%−4.40%+81.7
Apr-19 → Mar-205+52.70%+31.50%+21.2
Apr-20 → Mar-215+394.50%+18.40%+376.1
Apr-21 → Mar-227+25.72%+28.70%−3.0
Apr-22 → Mar-237+59.72%−18.10%+77.8
Apr-23 → Mar-245+357.82%+26.30%+331.5
Apr-24 → Mar-257+57.78%+25.00%+32.8

How to read. Each row is a single Apr-1 to next-Apr-1 holding period. “Mīzān return” is the equal-weight Concentrated portfolio's total return; “SPY return” is the S&P 500 ETF's total return over the matched window (locked Apr-Apr series, source mizan_sp500_apr_apr). “Excess” is Mīzān return minus SPY return in percentage points. The two highlighted rows (2020 and 2023 vintages) are the post-COVID and post-Inflation Recovery deployments — these triple-digit prints concentrate roughly 60% of the 29-year terminal NAV.

Rolling-window stability · Concentrated 5-7

Overlapping 3-year, 5-year, and 10-year holding periods anchored at every fiscal-year endpoint. Range expressed as best-window CAGR / worst-window CAGR. In this backtest, the worst 10-year stretch on the 29-year record (+48.8% CAGR) sat above SPY's best 10-year stretch (+16.6% CAGR) over the windows measured.

3-yr rolling CAGR · range
+30.5%+135.0%
26 overlapping 3-year windows (1999–2024). Best: 2018–2020, +135.0% CAGR. Worst: 2011–2013, +30.5% CAGR. SPY range: −14.6% (2000–2002) to +27.6% (1997–1999). Even Mīzān's worst 3-year stretch beat SPY's best 3-year stretch by roughly 3 percentage points.
5-yr rolling CAGR · range
+31.5%+134.0%
24 overlapping 5-year windows (2001–2024). Best: 2020–2024, +134.0% CAGR. Worst: 2011–2015, +31.5% CAGR. SPY range: −2.3% to +18.5%. In this backtest, the strategy's worst 5-year compounding rate (+31.5%) exceeded SPY's best 5-year rate (+18.5%) over the windows measured.
10-yr rolling CAGR · range
+48.8%+86.4%
19 overlapping 10-year windows (2006–2024). Best: 2014–2023, +86.4% CAGR. Worst: 2008–2017, +48.8% CAGR. SPY range: −1.4% (1999–2008) to +16.6% (2012–2021). The 10-year band tightens substantially — the strategy's compounding rate is much more stable at multi-year horizons than at single vintages.
Windows beating SPY
100% / 100% / 100%
Across the 26 three-yr, 24 five-yr, and 19 ten-yr overlapping windows in this backtest, each window outperformed SPY over its measured span. At annual frequency the record is 27 of 29 (the two underperforming years were 1998 and 2021, both by under 11 pp). These are in-sample, backtested windows — not a forward guarantee.
Best annual vintage
+394.5%
2020 vintage (Apr 2020 → Mar 2021) — the post-COVID Recovery deployment. SPY same period: +18.4%. Excess return: +376.1 pp. The 2023 vintage was the second-best at +347.5%.
Worst annual vintage
−0.57%
2008 vintage (Apr 2008 → Mar 2009) — Normal mode through the Lehman/early-GFC year. SPY same period: −37.0%. Excess return: +36.4 pp. The only negative year on the 29-year record — and it still beat the index by nearly 37 percentage points.
Monthly returns heatmap · Concentrated 5-7

348 monthly observations across 1996-04 → 2025-03 (plus the live partial period through 2025-Q1). Cell color saturates at ±15%; deeper green = stronger month; deeper red = weaker month. Scan vertically for seasonal patterns, horizontally for vintage patterns — the 2020-04 (+49.9%) and 2024-02 (+36.2%) cells are the two strongest months on record; 2020-03 (−26.3%) and 2018-10 (−21.9%) the two weakest.

Year Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec

Computation. Calendar-month returns derived from end-of-month NAV samples on the daily reconstruction (mizan_strategy_daily_nav, strategy_id = 1). April 1996 is partial (inception 1996-04-01); March 2025 is the latest closed month. Color encoding is linear in monthly return, capped at ±15% saturation.

Dual-mode selection.

Mīzān applies different filters depending on whether the prior calendar year was a broad drawdown. Specifics are proprietary; the behavioural profile of each mode is disclosed below. Across the eleven Recovery vintages on file, every single one closed positive — the worst at +26.5%. The single negative year in the entire 1996–2024 record was a Normal-mode vintage (2008, −0.57%).

Recovery Mode · 11 of 29 years

The rebound portfolio.

Deployed after market downturns. Emphasis on forward earnings momentum, filtered by balance-sheet strength. Thesis: earnings recoveries are systematically mispriced after crises.

Mean
+132.2%
Median
+97.7%
Best
+394.5%
Worst
+26.5%
Normal Mode · 18 of 29 years

The compounding portfolio.

Default mode. Emphasis on the gap between intrinsic value and market price, among quality-filtered names. Thesis: mispricing of future cash flows in calmer markets.

Mean
+54.5%
Median
+52.7%
Best
+139.9%
Worst
−0.57%

One rule. Three concentration levels.

The same selection logic can be deployed at different portfolio sizes. Each variant was backtested over the identical 29-year window with identical rules and universe. The concentrated 5–7 name configuration is the primary subject of this research and the one the author runs with personal capital. Broader configurations are shown for completeness and to illustrate how the rule scales.

Concentrated

Primary configuration
Portfolio size5–7 names
CAGR70.30%
Sharpe (annual)0.91
Sharpe (daily)1.72
Beta vs S&P1.04
Annual vol89.5%
Daily vol (annlsd)32.4%
Max drawdown−52.77%
Worst year−0.57%
Positive years28 / 29

The live configuration. Highest absolute return. Single-name risk is real — the safety filter exists for this reason.

Broad-20

Reference · smoother
Portfolio size20 / year
CAGR51.22%
Sharpe (annual)1.16
Sharpe (daily)1.66
Beta vs S&P0.78
Annual vol46.3%
Daily vol (annlsd)24.9%
Max drawdown−50.06%
Worst year−2.79%
Positive years28 / 29

Shown to demonstrate that the rule's edge does not vanish with diversification.

Broad-40

Reference · smoothest
Portfolio size40 / year
CAGR41.37%
Sharpe (annual)1.14
Sharpe (daily)1.52
Beta vs S&P0.71
Annual vol37.2%
Daily vol (annlsd)22.5%
Max drawdown−45.06%
Worst year−5.88%
Positive years28 / 29

Lowest single-name risk. The edge persists even at this breadth.

Pick-level distribution.

Across 181 individual one-year holdings, the aggregate hit-rate and return profile at the pick level — followed by the top and bottom ten.

Total picks
29 holding years · 5–7 names each.
Hit rate
Picks that closed positive over their April–April year.
Mean per pick
Arithmetic average across all 181 holdings.
Median per pick
The typical pick — half above, half below.

Top 10 winners single year, among 181 picks

Bottom 10 losers single year, among 181 picks

Every year. Every pick.

Click any row to expand the individual names held for that 1 April to 1 April period and their realised returns.

Year
Return
Picks
W
L
Best
Worst

Capacity is the binding constraint — and it scales.

The 70.30% theoretical CAGR is computed in a frictionless world. In the real one, capacity is governed by two readings of the cohort’s realised dollar volume at rebalance: the strict equal-weight reading (constrained by the marginal pick) and the cohort-aggregate reading (constrained by total cohort liquidity at the same participation rules). The first is the conservative ceiling; the second is the higher ceiling implied when position sizes are allowed to track per-name liquidity. Across all three configurations, the cohort-aggregate ceiling at moderate participation runs from $5B (Concentrated·7) to $57B (Broad-40). What follows is the era progression of the universe floor, the Year One cohort baselines for all three configurations, the portfolio-construction rules used to derive both readings, and the resulting capacity bands. These are research properties of the rule’s liquidity — not an offer or an invitation to deploy capital.

Universe floors by era

Mīzān’s minimum market-cap floor steps up by era to track the natural growth of the listed universe. These floors are part of the published methodology — the ranking and selection mechanics that follow them are not.

1990–1999
$50M
Pre-internet float, smaller listed universe.
2000–2009
$250M
Post dot-com. Mid-cap concentration grows.
2010–2019
$500M
Listed universe expands; small-cap thinning.
2020–present
$1B
Megacap dominance; higher liquidity threshold across the listed universe.

Era progression reflects the listed universe getting larger, not the rule getting permissive. A “mid-cap” in 2024 is structurally different from a “mid-cap” in 1996.

Year One reference baselines · all three configurations

The capacity arithmetic begins from real cohorts. Concentrated (7 names) is the deployed Year One book (1 April 2025); Broad-20 and Broad-40 are the most-recent backtest cohorts (1 April 2024 vintage), with all ADVs measured against the same 60-day pre-deployment window (Jan–Mar 2025) so the three configurations sit on a single, internally consistent liquidity reading. Two numbers per cohort matter: the marginal pick (smallest-by-ADV name — binds equal-weight capacity) and the cohort-aggregate ADV (sum across all names — binds capacity under ADV-proportional sizing).

Configuration
Names
Marginal pick · ADV
Cohort aggregate ADV
Concentrated · 7 namesLive Year One deployment, 2025-04-01 cohort.
7APP, GEV, CRDO, TOST, AS, ROAD, SEZL
SEZL · $40.9M$1.19B market cap; above the 2020s era floor of $1B.
$4.91B / dayLargest pick (APP $2.72B) is 66× the smallest.
Broad-20 · 20 namesBacktest cohort (2024-04-01); same selection logic, deeper rank cut.
20NVDA, APP, UBER, VST, FSLR, SFM, … DRVN
DRVN · $11.0MTop of the cohort: NVDA at $35.1B/day.
$42.3B / dayCohort dollar liquidity ~9× the Concentrated cohort.
Broad-40 · 40 namesBacktest cohort (2024-04-01); broadest published configuration.
38 eligibleTwo AKO ADR classes (Chilean bottler, <$1M ADV) substituted out.
IAS · $9.2MBelow IAS, the next-tier names absorb capacity proportionally.
$54.3B / dayIncludes NVDA $35B, PLTR $8.5B, APP $2.7B at the top.

Reading note. The Concentrated cohort is the live deployed book; Broad-20 and Broad-40 are backtest cohorts shown here at face value. All ADVs are computed against the same Jan–Mar 2025 window so the comparison is liquidity-consistent rather than vintage-consistent — a deliberate choice to answer the “what would deployment look like today?” question. The two AKO dual-class ADR positions in Broad-40 (Embotelladora Andina, total $0.2M combined ADV) fall below a $5M ADV liquidity floor and are substituted with the next-eligible candidates in any deployed configuration; the 38-name reading reflects that substitution.

Portfolio construction rules

The bands below are not picked from the air. They apply three standard liquidity constraints to the marginal pick’s realised volume. A reader can verify the arithmetic against any name in the cohort.

i.
Participation cap
Daily order flow capped at a fraction of the name’s ADV to keep market-impact bounded. Three regimes are tabulated below: 10% conservative, 15% moderate, 20% aggressive.
ii.
Build window
Number of trading days over which the position is built. A wider window absorbs more shares without lifting the price; the bands assume 5 / 7 / 10 days for the three regimes respectively.
iii.
Float ownership cap
Position capped at 5% of the company’s float to bound concentration risk. For the Year One cohort this constraint is non-binding — ADV-times-build dominates.

Equal-weight ceiling  ·  max position = participation% × marginal-pick ADV × build_days  ·  AUM = max position × N
Cohort-aggregate ceiling  ·  AUM = participation% × Σ ADV × build_days  ·  positions sized proportional to per-name ADV

Capacity bands · equal-weight ceiling

Strict equal-weight reading: every name carries the same dollar position, so the entire portfolio is constrained by the smallest-ADV pick. This is the conservative ceiling — the AUM at which the marginal pick still fits inside the participation cap without lifting price. Concentrated benefits at this ceiling because its rank cut sits high in the universe; the broader configurations carry a lower marginal-pick ADV and therefore a lower equal-weight ceiling per name, partially offset by holding more names.

Configuration
Conservative
10% × 5d
Moderate
15% × 7d
Aggressive
20% × 10d
Concentrated · 5 namesHighest alpha density. SEZL ($40.9M) as marginal across 5 holdings.
~$100M$20M / position
~$215M$43M / position
~$410M$82M / position
Concentrated · 7 namesLive Year One deployment. SEZL ($40.9M) marginal × 7.
~$145M$20M / position
~$300M$43M / position
~$575M$82M / position
Broad-20 · 20 namesDRVN ($11.0M) marginal × 20 holdings.
~$110M$5.5M / position
~$230M$11.5M / position
~$440M$22M / position
Broad-40 · 38 namesIAS ($9.2M) marginal × 38 holdings (after AKO substitution).
~$175M$4.6M / position
~$365M$9.7M / position
~$700M$18M / position
At strict equal-weight, the four configurations cluster between $100M and $700M depending on regime — capacity differences across configurations are second-order. The first-order question is what happens when position sizing is allowed to track per-name liquidity. That is the next table.
Capacity bands · cohort-aggregate ceiling

Liquidity-proportional reading: each name is sized within its own ADV-times-build participation cap, with the equal-weight target as a soft anchor. The implied capacity is then bounded by the cohort-aggregate dollar liquidity at the same regime parameters. The numbers below illustrate the scaling property of the rule: as the rank cut widens from 7 to 40 names, cohort dollar liquidity grows from $4.9B to $54.3B per day, and capacity scales linearly with that.

Configuration
Conservative
10% × 5d
Moderate
15% × 7d
Aggressive
20% × 10d
Concentrated · 7 names$4.91B daily cohort liquidity.
~$2.5B10% × $4.91B × 5d
~$5.2B15% × $4.91B × 7d
~$9.8B20% × $4.91B × 10d
Broad-20 · 20 names$42.3B daily cohort liquidity.
~$21B10% × $42.3B × 5d
~$44B15% × $42.3B × 7d
~$85B20% × $42.3B × 10d
Broad-40 · 38 names$54.3B daily cohort liquidity.
~$27B10% × $54.3B × 5d
~$57B15% × $54.3B × 7d
~$109B20% × $54.3B × 10d
The cohort-aggregate ceiling describes the dollar liquidity the rule’s cohort could in principle absorb. At moderate participation, the Concentrated cohort implies roughly $5B; Broad-20 roughly $44B; Broad-40 roughly $57B. The configurations sit at different points on the return–capacity surface: Concentrated concentrates per-name conviction, Broad-40 carries the most cohort liquidity, Broad-20 sits in between. The rule is the same across all three; capacity is a property of the liquidity, not an offer.
Capacity scaling · both ceilings, moderate participation (15% × 7d)

Logarithmic horizontal scale spanning $100M to $100B. Solid bars are the equal-weight ceiling (constrained by marginal pick); hatched extensions reach the cohort-aggregate ceiling (constrained by total cohort dollar liquidity). The gap between the two is the additional liquidity capacity that ADV-proportional sizing implies.

$100M $1B $10B $100B Concentrated · 7 EW $300M aggregate $5.2B Broad-20 EW $230M aggregate $44B Broad-40 EW $365M aggregate $57B Equal-weight ceiling · marginal-pick constrained Cohort-aggregate ceiling · ADV-proportional sizing

Reading. The equal-weight bars cluster between $230M and $365M — that is the rule run at strict simplicity, where every name carries the same dollar weight and the smallest pick caps the whole portfolio. The hatched extensions show the additional headroom that opens when each name is sized within its own ADV-times-build cap, summing across the cohort. Concentrated gains roughly 17×; Broad-20 gains roughly 190×; Broad-40 gains roughly 156×. These are the liquidity ceilings the rule implies at each reading; they describe a research property of the cohort, not a product on offer.

Operational realities
Holding period
12 months
No additions, no trims, no stop-losses. Single rebalance window per year on 1 April.
Annual turnover
~100%
Each April, the prior year’s portfolio is fully replaced by the new rank. A 5–10 day liquidation, 5–10 day deployment cycle is the realistic operational window.
Slippage assumption
10–40 bps / leg
Realistic per-leg slippage on 1 April rotation, scaling with AUM and concentration. At the upper bound of each band, slippage widens.
Tax drag (taxable account)
~25–40%
Annual full rotation generates short-term capital gains on every winning name. Pre-tax CAGR overstates after-tax compounding materially in the highest brackets.
Cohort depth (Normal mode)
8–35 names
Number of names that pass the published filter set in a typical screening year. The Concentrated configuration draws the top 5–7 from this cohort; depth in excess of seven represents the bench, not the deployed book.
Float-cap binding test
Non-binding
For the Year One cohort, the 5%-of-float constraint sits well above the ADV-times-build constraint at every regime. ADV is the binding wall — not ownership concentration.
The honest read. Capacity is two questions, not one. Question A: how much can a strict equal-weight deployment absorb before the marginal pick is binding? Answer: $100M–$700M depending on configuration. Question B: how much can a deployment sized to per-name ADV absorb before the cohort aggregate is binding? Answer: $5B (Concentrated) to $57B (Broad-40) at moderate participation. The first ceiling reflects the rule applied at strict equal weight; the second reflects positions sized to per-name liquidity. Both are research properties of the cohort’s realised volume — neither is an AUM on offer. The 70.30% theoretical CAGR is the rule’s gross return density — the slippage and tax drag any real dollar would pay scales with the regime. The honest reading prices return, capacity, and slippage separately.

Data provenance.

The figures on this page are reproducible from the dataset described below. Any number that cannot be reconciled to this provenance should be challenged.

Universe construction
US-listed common equities active on 1 April of each screening year, plus a delisted-companies set covering the modern record. The universe rebuilds annually.
Fundamental data
Vendor-supplied historical fundamentals at quarterly granularity, point-in-time where available. Where true point-in-time consensus was unavailable, current-estimate history is used and the limitation is disclosed in § 11.
Price data
Adjusted closing prices on 1 April (or next trading day where 1 April is a market holiday). Total return basis — dividends reinvested at ex-date.
Benchmark
S&P 500 total return index (SPX with dividend reinvestment), 1 April to 1 April windows identical to Mīzān’s holding period.
Holding-period convention
Each return is the simple price return from 1 April close to 1 April close of the following year, dividends reinvested. No mid-year marks; no intra-year drawdown observation.
Selection mechanics
Proprietary. Methodology disclosed structurally in § 01 and the universe floor by era is disclosed in § 09. Ranking weights, factor thresholds, and regime triggers are not published. The standard disclosure boundary for a private quantitative system.
Live verification
Year One picks were timestamped on 1 April 2025 and the closing returns are verifiable against public price history. The audit window opens with Year One on the live ledger.
Reproducibility
Every figure on this page is derived from the published return series and the provenance described above. The arithmetic is reproducible from those inputs.

What this backtest cannot tell you.

The honest limitations of any backtest. Reading these is a prerequisite to interpreting the numbers above.

Known structural limitations

Seven things you must account for.

No backtest, however carefully constructed, is a substitute for live results. The following are the material structural limitations of this one, as fairly as I can state them.

  • Survivorship bias. The universe is built from currently-listed companies plus a historical delisting set. Some delisted names from the 1990s–2000s may still be missing. This likely inflates returns modestly.
  • Look-ahead bias on forward estimates. Some inputs rely on analyst forecasts as they existed on 1 April of each screening year. Where true point-in-time consensus was unavailable, current-estimate history may introduce slight distortion.
  • Retrospective regime classification. Recovery vs Normal mode is classified using full calendar-year drawdown data. In live trading, regime could be determined with Q1 data only, but this has not been stress-tested in-sample.
  • No transaction costs or slippage. Returns assume perfect execution at April 1 closing prices. Live trading would incur bid-ask spread, commissions, and some market impact on smaller-cap names.
  • No tax drag. All returns are pre-tax. Annual turnover generates short-term capital gains that would meaningfully reduce net compounded returns in taxable accounts.
  • High single-year variance. Annual volatility of 89.5% is very high. Year-to-year outcomes can diverge dramatically from the long-run average. The mean is not a promise for any given year.
  • The path inside any year can be brutal. Intra-year drawdowns are now measured (see § 03) on a daily NAV reconstruction. The worst peak-to-trough loss on the Concentrated configuration was −52.77% over 27 calendar days during the February–March 2020 shock; full recovery took 61 more days. Annual reporting showed perf-year 2020 closing at +394.5%, a figure entirely disconnected from what holding the rule felt like in the spring of that year. A concentrated 5–7 name portfolio can sit substantially below its starting value for weeks or months inside any holding year, even when the year ultimately closes positive. The honest statement of risk is the −52.77% historical max drawdown alongside the 32.4% annualised daily volatility — not the 28 of 29 positive-year count, which describes outcomes at the year boundary, not the path inside it.
  • Concentrated portfolios carry stock-specific risk. Five to seven names means a single blow-up meaningfully impairs a year. The balance-sheet filters exist to minimise this, but the risk is not zero.
How to read the live experiment in light of these caveats: the live portfolio starting April 2025 uses the identical selection rules, with actual capital deployed by the author, and cannot be cherry-picked after the fact. Year One closed at +68.92% versus the S&P 500's +16.80% for the same window — above the rule's historical median (+59.71%), below the mean (+83.99% — pulled by the +394.5% recovery vintage of 2020). Inside the distribution, not above it. Year Two is now under way; a live record of follow-on years is what the live ledger exists to compile. Whatever the backtest's flaws, the live record corrects them in real time.