Carlo Hofer

← Volatility-managed portfolios

Investment research

Volatility-managed portfolios: the results

Volatility-managed factors scale each month's position by the inverse of last month's realised variance. Moreira and Muir (2017) and the real-time critique of Cederburg et al. (2020), replicated on Kenneth R. French's six equity factors, then the 136 months after Moreira and Muir's sample, May 2015 to August 2026. Every figure is a simulation drawn from the tables committed with the code.

Simulated on Kenneth R. French's data to August 2026, vintage 202609Last build: 13 November 2026

The six expectations, their verdicts and the study's own question.

After the sample
0 of 6
factors whose real-time combination has a Sharpe ratio above the real-time original's beyond its 95% interval, May 2015 to August 2026, half-widths 0.02 to 0.64, simulated
In the sample
1 of 6
factors whose volatility-managed Sharpe ratio is above the original's beyond its 95% interval: momentum, +0.51 (0.07 to 0.96). The alpha's t-statistic is above 1.96 for 3 of 6. To April 2015, simulated
Market alpha net of costs
4.03
% a year at 10 bp per unit of turnover, t 2.31, and 3.13 at 20 bp, t 1.81, to April 2015, simulated
Expectations met
4 of 6
fixed before any run that judged them. Not met: E2, 22 of 24 comparisons (French revised RMW), and E4, 2 of 6 factors (the alpha and the Sharpe ratio test answer different questions)

The 136 months after the sample cannot tell any of these differences from zero, and the sign of a Sharpe ratio difference on one sample says nothing about another.

  • After the sample, May 2015 to August 2026, no factor's real-time combination has a Sharpe ratio above the real-time original's by more than its interval allows. E5 is met: the median difference is +0.007, annualised, and the median half-width 0.213. These 136 months cannot tell the differences from zero. The market's volatility-managed Sharpe ratio is 0.55 against the original's 0.78, a difference of −0.23 with an interval of −1.12 to 0.66 (Figure 1).
  • The alpha and the Sharpe ratio answer different questions: the alpha tests what the volatility-managed factor adds to a holding of the original, and the Sharpe ratio difference tests it in the original's place. On Moreira and Muir's sample Ledoit and Wolf's test finds the volatility-managed Sharpe ratio above the original's for momentum alone, a difference of +0.51, annualised, with an interval of 0.07 to 0.96, while the alpha's Newey-West t-statistic is above 1.96 for the market, momentum and RMW (Figure 2). E4 is not met: the two agree in sign and significance for 2 of 6 factors where 5 were expected, though the test rejects 3.7 per cent of 1,000 simulated pairs with equal Sharpe ratios, inside 3.5 to 6.5.
  • The market's position changes by 0.75 of the factor a month on average, and its alpha net of 10 basis points per unit of that change is 4.03 per cent a year with a t-statistic of 2.31, and 3.13 with a t-statistic of 1.81 at 20 (Figure 3). The position's median is 0.94 and its 99th percentile 7.27, against Moreira and Muir's 0.93 and 6.39 (Figure 4). E6 is met, 17 of 17.
  • On each paper's own sample the replication matches Moreira and Muir's Table I for 6 of 6 factors (E1, the market's alpha 4.93 per cent a year with a White standard error of 1.57, against 4.86 and 1.56) and Barroso and Santa-Clara's scaled momentum in 4 of 4 comparisons (E3, a Sharpe ratio of 1.00 against 0.97 and a worst fall of 43.8 per cent against 45.2). Cederburg et al.'s results reproduce in 22 of 24 comparisons, and E2 is not met: the two misses are RMW's, whose history French has revised since Cederburg et al. downloaded it, its mean on their months now 3.12 per cent a year against their 2.92 (Table 8).

Table 1 | The six expectations, written before any run that judged them, and what the run found

Each expectation is a strict pass or fail, dated 8 or 10 November 2026 in src/sq/constants.py, and no level moved after a run.

ExpectationWhat was expectedWhat the run foundVerdict
E1Moreira and Muir's Table I on their sample: the alpha's sign and 5 per cent significance, the alpha within one published standard error, the appraisal ratio within 0.05, for at least five of six factors6 of 6 factors (Table 5)met
E2Cederburg et al.'s Tables 1 and 5 on their sample: four comparisons for each factor, Sharpe ratios within 0.05, the real-time differences with the published sign and significance22 of 24. RMW fails two, because French has revised its history since the paper: its mean on their months is now 3.12 per cent a year against their 2.92 (Table 8)not met
E3Barroso and Santa-Clara's scaled momentum, May 1927 to December 2011: both Sharpe ratios within 0.05, both maximum drawdowns within five points4 of 4 (Table 9)met
E4Ledoit and Wolf's test rejects 3.5 to 6.5 per cent of 1,000 simulated pairs with equal Sharpe ratios, and agrees with the alpha's Newey-West t-statistic in sign and significance for at least five of six factorsSize 3.7 per cent. 2 of 6 agree, because the alpha and the Sharpe ratio difference test different things (4 of 6 by category, in no verdict) (Figure 2)not met
E5May 2015 to August 2026: the real-time combination's Sharpe ratio above the real-time original's beyond the interval on at most two factors, and the median difference within the median half-width0 of 6, median +0.007 within 0.213 (Table 3)met
E6The market's alpha net of 10 basis points per unit of turnover more than two Newey-West standard errors above zero, and the positions' percentiles within their bands17 of 17. Net of 10 basis points 4.03 per cent a year, t 2.31, and net of 20 basis points 3.13, t 1.81 (Figure 3)met

Source: Kenneth R. French Data Library and the author's calculations. results/202609/e1_verdict_202609.csv to e6_verdict_202609.csv, notebooks 03 to 08.

Figure 1 | Sharpe ratio differences in the 136 months after Moreira and Muir's sample, May 2015 to August 2026

Annualised Sharpe ratio difference, with Ledoit and Wolf's 95% interval

FactorReal-time combination minus real-time original95% intervalHalf-widthp-valueVolatility-managed minus original95% intervalHalf-widthp-value
Mkt-RF−0.083−0.722 to 0.5560.6390.76−0.231−1.120 to 0.6590.8890.53
SMB+0.031−0.092 to 0.1540.1230.51−0.103−0.690 to 0.4850.5870.64
HML−0.090−0.234 to 0.0530.1440.18−0.315−1.032 to 0.4020.7170.30
Mom+0.198−0.342 to 0.7380.5400.38+0.177−0.449 to 0.8030.6260.49
RMW+0.052−0.229 to 0.3340.2820.71+0.140−0.394 to 0.6750.5350.57
CMA−0.018−0.040 to 0.0050.0220.11−0.257−1.046 to 0.5320.7890.43

These 136 months cannot tell any of the twelve differences from zero, and the sign of a difference on these months says nothing about another sample. E5 judges the first comparison, and the second carries no verdict. c is fixed on the months to April 2015, and the real-time combination uses only the months before each decision. Simulated, before costs.

Source: Kenneth R. French Data Library and the author's calculations. Monthly, May 2015 to August 2026, US equity factors, USD. results/202609/e5_real_time_tests_202609.csv and e5_volatility_managed_tests_202609.csv, notebook 07.

What the signal did in the 136 months after Moreira and Muir's sample, with the volatility-managed factor's scale fixed on the data to April 2015 and the real-time combination carried forward.

The study's own question is what the signal did after Moreira and Muir's sample ended. Both constructions run unchanged on May 2015 to August 2026: the volatility-managed factor's scale, c, is fixed on the months to April 2015, and the real-time combination uses only the months before each decision, so no parameter of either was set on these months. Figure 1 shows each difference with its interval. The intervals are wide because the months are 136: the half-widths of the volatility-managed comparisons, annualised, are 0.53 to 0.89 on these months and 0.21 to 0.44 on the 1,065 months of Moreira and Muir's sample (Figure 2). These months therefore cannot tell the differences from zero.

Table 2 | The four Sharpe ratios in the 136 months after the sample, May 2015 to August 2026

Annualised Sharpe ratio, simulated, before costs

FactorOriginalVolatility-managedReal-time combination (S1)Real-time original (S2)
Mkt-RF0.7840.5540.6930.775
SMB−0.199−0.301−0.180−0.211
HML−0.019−0.333−0.141−0.051
Mom0.2070.3830.3940.197
RMW0.2700.4100.3110.258
CMA−0.061−0.317−0.093−0.075

Each difference between two of these columns, with its interval, is in Figure 1.

Source: Kenneth R. French Data Library and the author's calculations. Monthly, May 2015 to August 2026. results/202609/after_sample_sharpe_202609.csv, notebook 07.

Table 3 | E5, the expectation for the months after the sample, and the same counts for the volatility-managed factor

Annualised Sharpe ratio differences

ComparisonFactors with the interval above zeroAllowedMedian differenceMedian half-widthVerdict
Real-time combination minus real-time original02+0.0070.213met
Volatility-managed minus original0−0.1670.672no verdict

Source: Kenneth R. French Data Library and the author's calculations. results/202609/e5_verdict_202609.csv and e5_volatility_managed_tests_202609.csv, notebook 07.

The three months with the largest gaps between the volatility-managed and the original market are October 2018, April 2020 and January 2018. In October 2018 the position was 3.56, and the market fell 7.65 per cent while the volatility-managed market fell 27.3 per cent. In April 2020 the position was 0.013, and the market rose 13.6 per cent while the volatility-managed market rose 0.18 per cent. In January 2018 the position was 3.38, and the market rose 5.58 per cent while the volatility-managed market rose 18.9 per cent.

What the volatility-managed factor adds to the original on Moreira and Muir's sample, how the Sharpe ratio test reads it, and what trading it costs.

The test of a Sharpe ratio difference is Ledoit and Wolf's (2008) studentised circular block bootstrap. Before it touches a factor it is run on 1,000 simulated pairs of 1,065 months whose Sharpe ratios are equal by construction, and it rejects 37 of them at the 5 per cent level, 3.7 per cent, inside the range of 3.5 to 6.5 fixed before the run.

Figure 2 | Sharpe ratio differences on Moreira and Muir's sample, each factor from its first month to April 2015

Annualised Sharpe ratio difference, volatility-managed minus original, with Ledoit and Wolf's 95% interval, on the axis of Figure 1

FactorAlpha, % a yeart (Newey-West)Sharpe ratio, managedSharpe ratio, originalDifference95% intervalp-valueCorrelation ρα/σ(1 − ρ) × Sharpe ratio, original
Mkt-RF4.932.800.5150.419+0.097−0.112 to 0.3050.330.601+0.2640.167
SMB−0.59−0.450.0840.227−0.143−0.362 to 0.0760.190.605−0.0530.090
HML1.731.360.3540.395−0.041−0.270 to 0.1890.720.539+0.1410.182
Mom12.836.240.9930.480+0.5140.072 to 0.9550.030.450+0.7770.264
RMW2.802.810.5980.407+0.192−0.108 to 0.4910.210.583+0.3610.170
CMA0.360.660.4140.538−0.124−0.337 to 0.0900.240.672+0.0520.176

The difference equals α/σ minus (1 − ρ) times the original's Sharpe ratio for every factor, to four decimals: the alpha measures what the volatility-managed factor adds to a holding of the original, and the difference measures it held in the original's place. The sign of a difference on these months says nothing about another sample. Simulated, before costs.

Source: Kenneth R. French Data Library and the author's calculations. Monthly, August 1926 (January 1927 for momentum, August 1963 for RMW and CMA) to April 2015. results/202609/sharpe_tests_mm_sample_202609.csv and alpha_and_sharpe_difference_202609.csv, notebook 06.

Figure 3 | The volatility-managed factor's alpha gross and net of trading costs, Moreira and Muir's sample

Alpha in % a year, with plus or minus two Newey-West standard errors

FactorTurnover, a monthAlpha, grosstNet of 10 bptNet of 20 bptBreak-even, bp
Mkt-RF0.754.932.804.032.313.131.8155
SMB0.55−0.59−0.45−1.25−0.94−1.92−1.42none
HML0.691.731.360.900.700.070.0621
Mom0.9512.836.2411.705.8210.565.37112
RMW0.622.802.812.062.091.321.3538
CMA0.570.360.66−0.32−0.60−1.01−1.865

Turnover is the absolute change in the position from the month before. The cost is charged per unit of it, and the factor's own rebalancing, the same for the managed and the original factor, is left out. Moreira and Muir's Table IV gives the market 3.98 per cent a year net of 10 basis points. SMB's gross alpha is negative and has no break-even cost. Simulated.

Source: Kenneth R. French Data Library and the author's calculations. Monthly, to April 2015. results/202609/costs_mm_sample_202609.csv, notebook 08.

Figure 4 | The volatility-managed market's position, August 1926 to April 2015

c over last month's realised variance, log scale

StatisticMoreira and Muir, Table VComputed
Median0.930.94
75th percentile1.591.62
90th percentile2.642.71
99th percentile6.397.27
Largest16.8, March 1964

The position is above 2 in 183 of the 1,065 months. For the market, a position of 7 holds seven times the capital in stocks, six of them borrowed at the bill rate. E6 asks the median, 75th and 90th percentiles within 0.15 of Moreira and Muir's and the 99th within 1.0.

Source: Kenneth R. French Data Library and the author's calculations. Monthly, August 1926 to April 2015. results/202609/volatility_managed_mm_sample_202609.csv and position_market_mm_sample_202609.csv, notebook 08.

Table 4 | The volatility-managed position on Cederburg et al.'s sample, published beside computed

c over last month's realised variance, with their definition of the realised variance

FactorMedian, publishedMedian, computed99th percentile, published99th percentile, computedLargest
Mkt-RF0.960.966.476.0013.1
SMB0.810.795.075.088.7
HML1.021.005.895.608.9
Mom1.011.018.648.3216.2
RMW1.111.095.025.278.7
CMA0.970.944.564.238.7

E6 asks each median within 0.15 and each 99th percentile within 1.0 of Cederburg et al.'s Table 1, Panel D.

Source: Kenneth R. French Data Library and the author's calculations. Monthly, to December 2016. results/202609/position_cowy_sample_202609.csv, notebook 08.

Each paper's published numbers beside the computed ones, on each paper's own sample.

Table 5 | Moreira and Muir's Table I on their sample, published beside computed

Alpha of the volatility-managed factor on its original, % a year

FactorMonthsAlpha, publishedAlpha, computedStandard error, publishedStandard error, WhiteStandard error, Newey-WestAppraisal ratio, publishedAppraisal ratio, computedE1 holds
Mkt-RF10654.864.931.561.571.760.330.33yes
SMB1065−0.58−0.590.910.911.31−0.07−0.07yes
HML10651.971.731.021.051.270.200.17yes
Mom106012.5112.831.711.722.050.860.87yes
RMW6212.442.800.830.891.000.420.44yes
CMA6210.380.360.670.670.550.080.07yes

The published standard errors are White's, as are the computed ones the verdict uses. The market's volatility-managed Sharpe ratio is 0.52 against Table V's 0.52.

Source: Kenneth R. French Data Library and the author's calculations. Monthly, each factor from its first month to April 2015. results/202609/e1_comparison_202609.csv, notebook 03.

Table 6 | Cederburg et al.'s Table 1 on their sample, published beside computed

Annualised Sharpe ratios and the Jobson-Korkie p-value of their difference

FactorOriginal, publishedOriginal, computedVolatility-managed, publishedVolatility-managed, computedDifference, publishedDifference, computedp-value, publishedp-value, computed
Mkt-RF0.420.4200.510.513+0.09+0.0920.300.31
SMB0.230.2260.080.076−0.15−0.1500.090.09
HML0.400.4010.380.367−0.02−0.0330.860.74
Mom0.480.4750.990.962+0.50+0.4870.000.00
RMW0.380.4060.510.566+0.13+0.1610.290.20
CMA0.530.5340.400.392−0.13−0.1420.230.19

E2's first clause asks both Sharpe ratios within 0.05. It fails for RMW alone, its volatility-managed Sharpe ratio 0.006 outside the band.

Source: Kenneth R. French Data Library and the author's calculations. Monthly, each factor from its first month to December 2016. results/202609/cowy_table_1_computed_202609.csv, notebook 04.

Table 7 | Cederburg et al.'s Table 5, Panel A, on their sample, published beside computed

Annualised Sharpe ratios of the real-time combination S1, the real-time original S2 and their differences

FactorS1, publishedS1, computedS2, publishedS2, computedS1 − S2, publishedS1 − S2, computedp-value, publishedp-value, computedS1 − S3, publishedS1 − S3, computed
Mkt-RF0.420.4150.460.458−0.04−0.0430.640.62−0.11−0.113
SMB0.140.1280.190.187−0.06−0.0590.370.36−0.12−0.122
HML0.380.3750.430.442−0.06−0.0670.410.33−0.12−0.122
Mom0.920.8890.490.477+0.44+0.4120.000.00−0.07−0.078
RMW0.440.5100.310.328+0.13+0.1820.530.38−0.14−0.123
CMA0.520.5260.560.561−0.03−0.0360.200.14−0.11−0.121

S3 is the ex post optimal combination. E2's second clause fails for RMW alone, in S1, 0.020 outside the band, and its third and fourth hold: S1 − S2 has the published sign for six of six factors and momentum's is the one significant difference.

Source: Kenneth R. French Data Library and the author's calculations. Monthly, each strategy from 120 months after the factor's first month to December 2016. results/202609/cowy_table_5_computed_202609.csv, notebook 04.

Table 8 | RMW's revision: the miss of E2 and two diagnostics computed after the verdict, which carry none

RMW, August 1963 to December 2016, 641 months

SeriesMean, % a yearStandard deviation, % a year
Cederburg et al.'s Table 12.927.71
The Internet Archive's copy of French's five-factor file, 9 July 20172.897.74
French's file, vintage 2026093.127.70

Moving RMW's monthly returns by one constant to Table 1's mean closes the volatility-managed Sharpe ratio's gap to Table 1 from 0.056 to 0.022, and S1's from 0.070 to 0.027. The archived file is fetched at run time and not committed.

Source: Kenneth R. French Data Library and the author's calculations. results/202609/e2_diagnostic_a_rmw_mean_moved_202609.csv and e2_diagnostic_b_rmw_archived_202609.csv, notebook 04.

Table 9 | Barroso and Santa-Clara's Table 3, May 1927 to December 2011, published beside computed

Momentum (WML) and scaled momentum, 1,016 months

StatisticWML, publishedWML, computedScaled WML, publishedScaled WML, computed
Mean, % a year14.4614.8616.5016.69
Standard deviation, % a year27.5327.4216.9516.77
Sharpe ratio0.530.540.971.00
Maximum drawdown, %−96.69−95.56−45.20−43.83

The paper's scaled series starts in March 1927. French's daily file of the ten momentum portfolios starts on 3 November 1926, so the first scaled month here is May 1927. E3 judges the two Sharpe ratios and the two maximum drawdowns.

Source: Kenneth R. French Data Library and the author's calculations. Monthly. results/202609/bsc_table_3_computed_202609.csv and e3_comparison_202609.csv, notebook 05.

The data, the constructions, what is known at each date, the costs, and what the study cannot show.

Method

Data and period. Kenneth R. French's daily and monthly files of the six factors and of the ten portfolios sorted on prior returns, vintage 202609, downloaded at run time and never committed.4 The market, SMB and HML start in July 1926, momentum in January 1927, RMW and CMA in July 1963, and every series is cut at August 2026. Each paper's results are compared on its own months: Moreira and Muir's to April 2015,1 Cederburg, O'Doherty, Wang and Yan's to December 2016,2 Barroso and Santa-Clara's from May 1927 to December 2011.3

The signal and the constructions. Each construction uses its paper's own realised variance. The volatility-managed factor holds c over last month's realised variance. The real-time combination holds, each month, the mean-variance weights of an investor with a risk aversion of 5 on the volatility-managed and the original factor, estimated from all the months before it once 120 are available, its total position capped at 5. Scaled momentum holds WML at 12 per cent a year over its realised volatility of the previous 126 trading days. Rebalancing is monthly throughout.

What is known at each date. Each month's position uses only the realised variance of the month before. c is the exception: Moreira and Muir set it on the whole sample, which no investor could have known, and the replication does the same, as the paper does. The real-time combination uses only the months before each decision, and for the months after the sample c is fixed on the months to April 2015, so nothing in those 136 months was set on them.

The test. Ledoit and Wolf (2008) take the standard error of a Sharpe ratio difference by the delta method from the prewhitened Parzen kernel, and its interval from a circular block bootstrap of 2,000 samples of 12-month blocks, each sample's difference studentised by its own standard error.5 The paper proposes the QS kernel and reports the Parzen kernel's performance as very similar, and it calibrates the block length from the data, where the study fixed it at 12 months before the run.

Costs. A month's turnover is the absolute change in the position from the month before. The cost, 10 basis points per unit of turnover and 20 as the sensitivity, is subtracted from the volatility-managed return, and the alpha is estimated again on the net return. The factor's own rebalancing is the same for the managed and the original factor and is left out.

What the study cannot show. It holds no view on expected returns and estimates no premium. Its costs are charged on the change in a factor's position, and a book that holds the stocks pays more, so the break-even costs bound what it could pay. French revises history, so a run on a later vintage need not reproduce every digit, as RMW shows. The 136 months after the sample are too few to tell any Sharpe ratio difference from zero: halving the market's half-width would require four times as many. The volatility-managed position is never capped, as in the paper, so its largest positions, up to 16.8, are positions no mandate would allow.

Terms

Factor
the monthly return of a long-short portfolio that captures one source of common movement in stock returns, as Kenneth French publishes it.
Excess return
a return minus the one-month Treasury bill rate. Every factor is one.
Realised variance
the variance of a factor computed from its daily returns within one month, each paper with its own definition.
The signal
last month's realised variance of a factor, and nothing else.
Volatility-managed
Moreira and Muir's construction, the factor's return times c over last month's realised variance. Managed describes the construction and says nothing about its performance.
c
the constant that gives the volatility-managed factor the original's standard deviation over the whole sample, so it is known only at the sample's end.
Position
the multiple of the factor held in a month. The papers call a position above 1 leverage.
Real-time combination (S1)
Cederburg et al.'s construction, the mean-variance weights on the volatility-managed and the original factor with a risk aversion of 5, estimated each month from the months before it, its total position capped at 5.
Real-time original (S2)
the original factor alone at its real-time mean-variance weight.
Ex post optimal combination (S3)
the mix of the two with the weights that were best over the whole sample, which no investor could have known.
Scaled momentum
Barroso and Santa-Clara's construction, momentum (WML, winners minus losers) times 12 per cent a year over its realised volatility of the previous 126 trading days.
Alpha
the intercept of the regression of the volatility-managed factor on its original, in per cent a year, always with its standard error or t-statistic.
Standard error
the standard deviation of an estimate across the samples it could have come from. White's allows the residual's variance to change, and Newey and West's at 12 lags also allows a month's residual to be correlated with the 12 before it.
Sharpe ratio
the mean excess return over its standard deviation, annualised.
Jobson-Korkie test
Cederburg et al.'s test that two return series over the same months have the same Sharpe ratio, which assumes normal and independent months.
Delta method
the standard error of a function of means, from the function's slopes and the means' covariance.
Appraisal ratio
the alpha over the residual's standard deviation, annualised.
Interval
the 95 per cent interval of Ledoit and Wolf's test for the true Sharpe ratio difference. The test rejects when it excludes zero.
Half-width
the distance from a difference to either end of its interval.
p-value
the probability of a difference at least as far from zero as the one computed, if the two true Sharpe ratios were equal.
Size
the share of samples in which a test rejects when the null is true.
Turnover
the absolute change in the position from the month before, the first month carrying none.
Basis point
a hundredth of a percentage point.
Break-even cost
the cost per unit of turnover at which the alpha falls to zero.
Maximum drawdown
the largest fall from a peak of the compounded return.
Band
the distance from a published number within which a computed one counts as matching, fixed before the code.
Verdict
whether an expectation is met. A count reported in no verdict is shown without one.
Vintage
the version of French's files, named by the CRSP release they were built from.

Notes

  1. Moreira, A. and Muir, T. (2017). Volatility-managed portfolios. Journal of Finance, 72(4), 1611-1644. https://doi.org/10.1111/jofi.12513
  2. Cederburg, S., O'Doherty, M. S., Wang, F. and Yan, X. S. (2020). On the performance of volatility-managed portfolios. Journal of Financial Economics, 138(1), 95-117. https://doi.org/10.1016/j.jfineco.2020.04.015
  3. Barroso, P. and Santa-Clara, P. (2015). Momentum has its moments. Journal of Financial Economics, 116(1), 111-120. https://doi.org/10.1016/j.jfineco.2014.11.010
  4. Kenneth R. French, Data Library, files built from the 202609 CRSP database.
  5. Ledoit, O. and Wolf, M. (2008). Robust performance hypothesis testing with the Sharpe ratio. Journal of Empirical Finance, 15(5), 850-859. https://doi.org/10.1016/j.jempfin.2008.03.002

Other projects

Optimal versus naive diversification. Under an index-tracking mandate on French's 49 industries, five covariance estimators understate their own tracking error, by 22 to 61 per cent. The replication of DeMiguel, Garlappi and Uppal (2009) reproduces 29 of the 30 Sharpe ratios that carry a band. Page, dashboard, note, code

Climate risk as an equity signal. An ownership graph linking 328 listed companies to the 5,115 tracked assets they own. Of 19 apparent ownership changes checked by hand, 12 had no transaction behind them. Page, note, code

Stock-level index tracking: a point-in-time benchmark, a long-only mandate, and its performance attribution by sector. A point-in-time S&P 500 universe from CRSP and Compustat, a long-only mandate held close to its cap-weighted benchmark, and an attribution of its active return by sector. In build: its findings are stated when the study is complete. Page, code

Rules-based portfolio. Runs a real-money portfolio under rules published before the first order: two UCITS ETFs at 70/30, a monthly cycle with pre-trade checks, look-through exposure and ex-ante risk on a public dashboard. Page, dashboard, guide, code