One thematic portfolio, three ways of managing it, and what fifteen years of history can and cannot say about each.
Summary
Kaleidoscope is a portfolio of 26 thematic ETFs held at fixed weights. Around it we run three managed versions. Each one answers a different question. Risk-Managed decides how much of the portfolio to hold. Selection Tilt decides which themes to hold more of. Tilt + Risk-Managed does both. This paper sets the four side by side over the same 15.2 years, from July 2011 to 30 September 2026, against an equal-weighted basket of the same themes and the S&P 500.
- The base portfolio is a higher-beta equity position. Kaleidoscope compounded at 16.9% a year against 14.1% for the S&P 500. It did so with a beta of 1.19, a 0.88 correlation to the index, and a maximum drawdown of −39.0%.
- Risk-Managed converts risk; it does not add return. It earned 17.9% a year, about a point more than the portfolio it manages, at 18.4% volatility rather than 23.1%. Its worst fall was −29.5% rather than −39.0%. It beat the base portfolio in only 4 of 14 full calendar years. It finishes ahead by losing less in the bad ones.
- Selection Tilt ranks themes, and the ranking works in the tails. Themes in the top fifth on six-month relative momentum beat the average theme by 0.73% a month. Those in the bottom fifth lagged by 0.60%. The middle three were indistinguishable from noise. The tilt earned 18.8% a year, but one year, 2017, carries much of that. The edge was clearly smaller in the second half of the record.
- Tilt + Risk-Managed had the best full-period record: 19.2% a year, a Sharpe ratio of 1.02 and a −29.8% maximum drawdown. Since the theme list was fixed in January 2022, it has the lowest return of the four, at 15.0% a year. In practice it is Risk-Managed plus a bet on the tilt.
- None of these improvements is statistically significant at conventional levels. The overlay’s rotation p-value is 0.104, the tilt’s 0.058 and the combination’s 0.369. All three rules were frozen on 28 September 2026 and are now being recorded forward.
Table 1. The four models and two benchmarks, 5 July 2011 to 30 September 2026
| Annual return | Volatility | Sharpe | Max drawdown | Beta to S&P 500 | $1 grew to | |
|---|---|---|---|---|---|---|
| Kaleidoscope | 16.9% | 23.1% | 0.79 | −39.0% | 1.19 | 10.7 |
| Risk-Managed | 17.9% | 18.4% | 0.99 | −29.5% | 0.75 | 12.3 |
| Selection Tilt | 18.8% | 22.7% | 0.87 | −35.7% | 1.12 | 13.7 |
| Tilt + Risk-Managed | 19.2% | 19.1% | 1.02 | −29.8% | 0.71 | 14.5 |
| Thematic equal weight | 14.6% | 21.9% | 0.73 | −40.4% | 1.12 | 7.9 |
| S&P 500 (SPY) | 14.1% | 17.1% | 0.86 | −33.7% | 1.00 | 7.4 |
The four models
All four hold the same 26 themes, and all four are measured the same way: daily, after a 15bp trading cost, with no management fee. They differ only in the decisions they make.
Kaleidoscope: the base portfolio
Kaleidoscope holds 26 thematic ETFs in four sections at fixed target weights. Technology is 40% of the portfolio, Consumer 30%, Commodities 15% and Industrial 15%. Individual weights run from 1% to 10%; the largest is the artificial intelligence fund AIQ. The portfolio rebalances to target each quarter and drifts in between. Because the targets never change, its only trading is the drift correction, about 19.1% of the portfolio a year.
Table 2. Kaleidoscope’s four sections and target weights
| Section | Target weight | Themes |
|---|---|---|
| Commodities | 15% | NANR 5%, COPX 3%, GDX 3%, MLPX 2%, URA 2% |
| Industrial | 15% | GRID 3%, ARKX 3%, PAVE 3%, IGF 3%, TAN 2%, PHO 1% |
| Technology | 40% | AIQ 10%, ARKQ 5%, CIBR 5%, SKYY 5%, SMH 5%, KWEB 5%, FDN 5% |
| Consumer | 30% | XBI 5%, ARKF 5%, GBTC 5%, BKCH 5%, ITB 3%, DRIV 3%, ESPO 2%, PEJ 2% |
Read as a whole, the portfolio is an equity position with more beta than the market, not a diversified collection of independent bets. Over the 15.2 years its beta to the S&P 500 was 1.19. It captured 1.20 times the index’s rising months and 1.20 times its falling months. Its daily correlation with the index was 0.88. That is the context for everything that follows: the managed versions are attempts to keep the themes’ higher return while paying less for their higher risk.
Risk-Managed: deciding how much to hold
Risk-Managed holds the Kaleidoscope portfolio and moves part of it to defensive assets when the portfolio’s own trend breaks. The rule is short:
- Signal. The portfolio’s value against its 125-day moving average, checked daily.
- Band. It de-risks when the portfolio falls 5% below the average and re-risks when it recovers 5% above it. Between those edges it holds its last position.
- Execution. The signal is acted on weekly, on every fifth session, the day after it is read.
- Size. When risk-off, 60% of the portfolio moves into the defensive sleeve; the other 40% stays in the themes.
- Sleeve. The sleeve holds long Treasuries (TLT) or gold (GLD), chosen by the level of five-year inflation expectations. It moves to Treasuries below 1.8%, to gold above 2.6%, and holds its choice in between.
The sleeve’s choice matters because the two assets protect against different things. Treasuries protect in a disinflationary drawdown such as 2008 or 2020. In an inflationary one, such as 2022, they fall with equities and gold does the work. Across 18 years the regime switch has happened only twice.
Over the common window the rule spent 22.4% of days partly out of the market, in 10 separate spells. Its turnover over the window was 79% of the portfolio a year. It earned 17.9% a year against the base portfolio’s 16.9%, with volatility of 18.4% against 23.1%. Its Sharpe ratio was 0.99 against 0.79. Its beta to the S&P 500 fell to 0.75, and it captured 0.75 of the index’s falling months while keeping 0.99 of its rising ones.
The calendar years show how that happens. Risk-Managed finished ahead of the base portfolio in only 4 of 14 full years. In strong years it gives up return: it trailed by 14.1 points in 2023, a year that began with the sleeve still on. In bad years it loses much less: −18.3% against −32.7% in 2022. Compounding favours the second effect. This is a portfolio for an investor who wants the themes’ exposure without their full drawdowns, not one that expects to beat the themes every year.
The longer history makes the same point more sharply. The overlay can be run from September 2008, though only nine of the 26 themes existed then. Over those 18.0 years Risk-Managed compounded at 18.4% against the base portfolio’s 16.1%. Its maximum drawdown was −29.5% against −42.7%, and a dollar grew to 21.0 rather than 14.6. Table 3 shows the five stress episodes in that record. The overlay lost less than the portfolio in every one, but it made money in only one of them.
Table 3. Risk-Managed in the five stress episodes since Sep 2008
| Episode | Risk-Managed | Kaleidoscope | Thematic EW | S&P 500 |
|---|---|---|---|---|
| GFC 2008 | −8.0% | −37.3% | −45.0% | −43.4% |
| Euro 2011 | 0.8% | −19.4% | −27.5% | −16.3% |
| Q4 2018 | −11.6% | −23.9% | −19.7% | −18.7% |
| COVID 2020 | −21.8% | −32.6% | −36.0% | −33.4% |
| 2022 bear | −23.8% | −35.5% | −32.8% | −24.1% |
Selection Tilt: deciding which themes to hold
The tilt starts from an equal-weighted book of the 26 themes and leans it toward the themes that have been working. Each month every theme is scored on its six-month return relative to the S&P 500, skipping the most recent month. The themes are then split into thirds:
- Top third: twice the equal weight.
- Middle third: equal weight.
- Bottom third: nothing.
There is no market timing and no cash. The tilt rebalances monthly, with turnover of about 188% a year.
The design follows directly from what the signal can and cannot see. Cut into fifths each month, the top fifth of themes beat the average theme by 0.73% a month and the bottom fifth lagged by 0.60%. The three middle fifths came in at +0.06%, −0.12% and −0.02%: no usable information. So the tilt trades only the tails and leaves the middle alone. Avoiding the bottom third matters almost as much as owning the top third. That makes the tilt as much a fade of losing themes as a bet on winning ones.
Over the common window the tilt earned 18.8% a year against 14.6% for equal weight and 16.9% for the fixed-weight Kaleidoscope portfolio. Its volatility was 22.7% and its maximum drawdown −35.7%. Measured against its own equal-weight book in the research record, it added 4.11 percentage points a year, with tracking error of 7.35% and an information ratio of 0.56. The average rank correlation between its monthly scores and the following month’s returns was 0.0665, positive in 59% of months.
Two features of that record need to be stated plainly.
- One year carries much of it. In 2017 the tilt beat equal weight by 38.4 points. Its median year added 1.5 points. Without 2017, the average year added 2.4 points rather than 5.0.
- The edge was smaller recently. In the first half of the research record the tilt added 6.6 points a year; in the second half, 1.5. It trailed equal weight in 2019 and 2023. Its result for 2024 through 2026 so far is close to zero.
On 30 September 2026 the tilt holds 18 of the 26 themes. Eight are at twice the equal weight: AIQ, ARKF, CIBR, FDN, GBTC, SKYY, SMH and XBI. Ten are at equal weight. Eight are not held: ARKQ, GDX, IGF, ITB, KWEB, PHO, TAN and URA.
Tilt + Risk-Managed: both decisions at once
The fourth model holds the Selection Tilt and applies the Risk-Managed overlay to it. The overlay reads the base Kaleidoscope portfolio’s trend, not the tilt’s, so one risk signal governs both managed books. When the overlay is risk-off, 60% of the tilted book moves into the same Treasury-or-gold sleeve.
Over the full window it is the strongest of the four. It earned 19.2% a year at 19.1% volatility, a Sharpe ratio of 1.02, with a −29.8% maximum drawdown. Its beta to the S&P 500 was 0.71; it captured 1.03 of the index’s rising months and 0.74 of its falling ones. Against the tilt alone, the overlay cut the beta to 0.737 and volatility to 0.838 of the tilt’s.
It is not a strictly better version of Risk-Managed. Its daily correlation with Risk-Managed is 0.89. It finished ahead of Risk-Managed in 6 of 14 full years. Its worst year against Risk-Managed was 2023, when it trailed by 12.8 points. Since January 2022 it has compounded at 15.0% a year against Risk-Managed’s 17.5%. The honest description is Risk-Managed plus a bet on the tilt. It suits an investor who wants both decisions and accepts that the second one adds its own risk.
Side by side
Calendar years
Table 4. Calendar-year returns
| Year | Kaleidoscope | Risk-Managed | Selection Tilt | Tilt + RM | Thematic EW | S&P 500 |
|---|---|---|---|---|---|---|
| 2011 (from Jul) | −8.9% | 5.0% | −15.5% | 1.1% | −18.5% | −5.2% |
| 2012 | 21.4% | 18.0% | 18.7% | 15.8% | 12.9% | 16.0% |
| 2013 | 31.3% | 31.3% | 32.0% | 32.0% | 20.8% | 32.3% |
| 2014 | 18.1% | 10.4% | 5.9% | 0.6% | 4.4% | 13.5% |
| 2015 | 8.0% | 3.8% | 1.3% | −6.0% | −3.4% | 1.3% |
| 2016 | 5.8% | 14.0% | 18.7% | 24.1% | 19.9% | 12.0% |
| 2017 | 37.1% | 37.1% | 102.2% | 102.2% | 63.8% | 21.7% |
| 2018 | −8.9% | 1.3% | −18.5% | −9.2% | −18.5% | −4.6% |
| 2019 | 35.5% | 22.0% | 25.6% | 14.7% | 36.2% | 31.2% |
| 2020 | 42.9% | 39.7% | 71.3% | 70.6% | 55.0% | 18.4% |
| 2021 | 14.3% | 14.3% | 15.1% | 15.1% | 13.6% | 28.7% |
| 2022 | −32.7% | −18.3% | −19.8% | −13.5% | −28.8% | −18.2% |
| 2023 | 52.3% | 38.2% | 35.7% | 25.3% | 42.2% | 26.2% |
| 2024 | 25.2% | 23.1% | 20.9% | 18.9% | 22.0% | 24.9% |
| 2025 | 33.7% | 35.4% | 34.3% | 35.1% | 33.4% | 17.7% |
| 2026 (to Sep) | 15.5% | 13.8% | 12.8% | 11.3% | 12.0% | 12.7% |
Stress episodes
The deepest five falls of the base portfolio in the common window, measured from its peak to its trough, and what each model did over the same dates:
Table 5. The base portfolio’s five deepest falls in the common window, peak to trough
| Peak to trough | Kaleidoscope | Risk-Managed | Selection Tilt | Tilt + RM | Thematic EW | S&P 500 |
|---|---|---|---|---|---|---|
| Jul 2015 – Feb 2016 | −28.6% | −15.8% | −22.5% | −13.4% | −21.0% | −11.7% |
| Jul 2018 – Dec 2018 | −26.5% | −14.7% | −22.3% | −10.5% | −23.0% | −15.7% |
| Feb 2020 – Mar 2020 | −34.7% | −29.5% | −35.4% | −29.8% | −38.0% | −29.1% |
| Nov 2021 – Oct 2022 | −39.0% | −28.0% | −35.7% | −29.8% | −40.2% | −22.6% |
| Feb 2025 – Apr 2025 | −23.5% | −15.4% | −22.9% | −15.9% | −22.3% | −18.6% |
The two overlaid models lost less than the base portfolio in all five episodes. The tilt alone lost less in four and roughly the same in the COVID crash. The overlay did least in COVID, which fell too fast for a 125-day average to react. It did most in the slower declines of 2015 to 2016 and late 2018.
How the four relate
Table 6. Correlation of daily returns
| Kaleidoscope | Risk-Managed | Selection Tilt | Tilt + RM | Thematic EW | S&P 500 | |
|---|---|---|---|---|---|---|
| Kaleidoscope | 1.00 | 0.87 | 0.91 | 0.76 | 0.94 | 0.88 |
| Risk-Managed | 0.87 | 1.00 | 0.80 | 0.89 | 0.80 | 0.70 |
| Selection Tilt | 0.91 | 0.80 | 1.00 | 0.88 | 0.95 | 0.84 |
| Tilt + RM | 0.76 | 0.89 | 0.88 | 1.00 | 0.80 | 0.64 |
| Thematic EW | 0.94 | 0.80 | 0.95 | 0.80 | 1.00 | 0.88 |
| S&P 500 | 0.88 | 0.70 | 0.84 | 0.64 | 0.88 | 1.00 |
The base portfolio and the tilt move closely together (correlation 0.91), as do Risk-Managed and Tilt + Risk-Managed (0.89). The overlay is what separates the families. It lowers each book’s correlation to the S&P 500, from 0.88 to 0.70 for the base portfolio and from 0.84 to 0.64 for the tilt.
Since the themes were chosen
The 26 themes were selected at the end of 2021, by people who could see how the preceding years had gone. Everything before January 2022 is therefore a backcast. It shows how today’s themes would have behaved, not how a portfolio chosen in 2011 would have. The only period free of that hindsight is January 2022 onward, 4.7 years that include the 2022 bear market.
Table 7. Since the themes were chosen: 3 January 2022 to 30 September 2026
| Annual return | Volatility | Sharpe | Max drawdown | Beta to S&P 500 | |
|---|---|---|---|---|---|
| Kaleidoscope | 15.6% | 25.7% | 0.69 | −37.0% | 1.30 |
| Risk-Managed | 17.5% | 20.5% | 0.89 | −25.6% | 0.88 |
| Selection Tilt | 15.7% | 24.1% | 0.73 | −29.6% | 1.18 |
| Tilt + Risk-Managed | 15.0% | 20.6% | 0.78 | −23.2% | 0.82 |
| Thematic equal weight | 13.9% | 24.1% | 0.66 | −34.3% | 1.23 |
| S&P 500 (SPY) | 12.0% | 17.3% | 0.74 | −24.5% | 1.00 |
In this period the ordering changes. Risk-Managed leads at 17.5% a year with the second-smallest drawdown, −25.6%. Tilt + Risk-Managed had the smallest drawdown, −23.2%, but the lowest return of the four models, 15.0%. The tilt alone barely differed from the base portfolio, 15.7% against 15.6%. All four beat the S&P 500’s 12.0% a year, but the base portfolio did so with a −37.0% drawdown against the index’s −24.5%.
What the evidence supports, and what it does not
What it supports. Over every window measured, the overlay turned the portfolio’s equity-like risk into something closer to a balanced one. It cut beta, volatility and the worst losses, at a cost of return in strong years that compounding more than repaid. That conclusion rests on a mechanism, on consistency across five stress episodes, and on similar results in both halves of the record. Its Sharpe ratio was 1.06 in the first half and 0.96 in the second.
What it does not. None of the three improvements clears a 5% significance test:
Table 8. Statistical tests of each improvement
| Claim | Measured on | Rotation p-value | Forward years to resolve |
|---|---|---|---|
| The overlay improves the base portfolio’s risk-adjusted return | Base portfolio, since 2008 | 0.104 | – |
| Momentum ranks the themes (tilt beats equal weight) | Tilt vs equal weight, since 2011 | 0.058 (one-sided null: 0.068) | 13.3 (rank correlation), 19.8 (active return) |
| The overlay improves the tilt | Tilt + RM vs tilt, since 2011 | 0.369 | 3.4 (beta), 4.5 (volatility) |
These tests rotate each rule’s decisions against the returns many times and ask how often chance alone does as well. A p-value of 0.104 means a time-shifted copy of the overlay’s decisions did as well as the real ones about one time in ten. The tilt comes closest, at 0.058, and its record leans on a few large years. The combination, at 0.369, is not distinguishable from chance. Its case is that it inherits the overlay’s risk reduction, not that the two decisions add up.
Why the history cannot settle it. Thematic ETFs are young. Only 12 of the 26 themes existed when the tilt’s record begins in 2011, and only nine in 2008. The rules were designed against this same history, so the backtest is in-sample by construction. It also overlaps the hindsight in the theme list described above. At the precision these statistics have, confirming the tilt’s rank correlation would take about 13.3 years of new data, and its active return about 19.8 years. Measuring the overlay’s effect on the tilt’s beta and volatility would take about 3.4 and 4.5 years, and only if the market supplies at least four risk-off spells in that time.
What happens next. Each rule was frozen on 28 September 2026: its parameters, its code and its in-sample results are fixed and recorded. From that date every decision is logged as it is made. The forward record is built to reject a rule that is actively harmful, not to confirm a good one quickly. A forward result close to the recent past, such as the tilt adding about 1.5 points a year, is the expected outcome rather than a failure.
Where the models stand
At 30 September 2026:
- Risk-Managed is fully invested. The portfolio sits 2.9% above its 125-day average, inside the band. The last change was a re-risk effective 22 April 2026. The inflation gauge reads 2.36%, so a new risk-off signal would go to gold.
- Selection Tilt is overweight AIQ, ARKF, CIBR, FDN, GBTC, SKYY, SMH and XBI. It holds no ARKQ, GDX, IGF, ITB, KWEB, PHO, TAN or URA.
- Tilt + Risk-Managed holds the same tilted book, fully invested.
- Kaleidoscope holds its fixed weights, reset at the 30 September rebalance.
Table 9. Holdings at 30 September 2026
| Theme | Section | Kaleidoscope | Risk-Managed | Selection Tilt | Tilt + RM |
|---|---|---|---|---|---|
| NANR | Commodities | 5.0% | 5.0% | 3.8% | 3.8% |
| COPX | Commodities | 3.0% | 3.0% | 3.8% | 3.8% |
| GDX | Commodities | 3.0% | 3.0% | – | – |
| MLPX | Commodities | 2.0% | 2.0% | 3.8% | 3.8% |
| URA | Commodities | 2.0% | 2.0% | – | – |
| ARKX | Industrial | 3.0% | 3.0% | 3.8% | 3.8% |
| GRID | Industrial | 3.0% | 3.0% | 3.8% | 3.8% |
| IGF | Industrial | 3.0% | 3.0% | – | – |
| PAVE | Industrial | 3.0% | 3.0% | 3.8% | 3.8% |
| TAN | Industrial | 2.0% | 2.0% | – | – |
| PHO | Industrial | 1.0% | 1.0% | – | – |
| AIQ | Technology | 10.0% | 10.0% | 7.7% | 7.7% |
| ARKQ | Technology | 5.0% | 5.0% | – | – |
| CIBR | Technology | 5.0% | 5.0% | 7.7% | 7.7% |
| FDN | Technology | 5.0% | 5.0% | 7.7% | 7.7% |
| KWEB | Technology | 5.0% | 5.0% | – | – |
| SKYY | Technology | 5.0% | 5.0% | 7.7% | 7.7% |
| SMH | Technology | 5.0% | 5.0% | 7.7% | 7.7% |
| ARKF | Consumer | 5.0% | 5.0% | 7.7% | 7.7% |
| BKCH | Consumer | 5.0% | 5.0% | 3.8% | 3.8% |
| GBTC | Consumer | 5.0% | 5.0% | 7.7% | 7.7% |
| XBI | Consumer | 5.0% | 5.0% | 7.7% | 7.7% |
| DRIV | Consumer | 3.0% | 3.0% | 3.8% | 3.8% |
| ITB | Consumer | 3.0% | 3.0% | – | – |
| ESPO | Consumer | 2.0% | 2.0% | 3.8% | 3.8% |
| PEJ | Consumer | 2.0% | 2.0% | 3.8% | 3.8% |
Methodology
Returns. Daily total returns from FactSet, compounded. All four models are charged 15bp on every trade. No management fee, tax or financing cost is deducted. Annual figures compound daily returns. Sharpe ratios use a zero risk-free rate. Capture ratios use monthly returns.
The base portfolio. Kaleidoscope’s history is rebuilt from its fixed target weights and the funds’ daily returns. Holdings drift with prices and reset to target on each quarterly rebalance date. Before every theme existed, the weights are spread over the themes that did.
Benchmarks. Thematic equal weight holds every Kaleidoscope theme that has started trading, at equal weight, reset monthly, with no costs. S&P 500 is SPY, total return.
Common window. Tables and exhibits cover 5 July 2011 to 30 September 2026 unless they say otherwise. The start is set by the tilt, which needs 12 themes with six months of history before it can rank them. The 2011 and 2026 columns of the calendar table are partial years.
Significance. Each p-value comes from a rotation test, which shifts the model’s decisions against the return series and records how often a misaligned copy does as well. All are two-sided except where the table notes otherwise.
Important information
The performance shown is hypothetical and backtested. It was produced by applying rules to historical data and does not represent actual trading or the results of any account. Before January 2022 it reflects themes selected later, with knowledge of that history. The rules were developed on the same history they are tested on. Backtested results have inherent limitations and do not reflect the impact of fees, taxes, market impact beyond the assumed trading cost, or the decisions a manager might have made under real conditions. Past performance, actual or hypothetical, is not a guide to future results. Thematic ETFs can be volatile and concentrated, and some of the funds named hold digital assets. This paper is for information only and is not an offer, a solicitation or a recommendation to buy or sell any security.
