When a smaller position looks like diversification - Technical article accompanying Portfoliothese | Issue 3

What a new strategy contributes beyond the reduction of an existing position: A case study from digital assets

PART A · FOR PRACTICE

The problem

Anyone who takes a new strategy into a portfolio has to reduce something else to make room for it. If the result is then compared with the previous portfolio, the new strategy is also credited with the effect that arises solely from the smaller legacy position. How large this share can be is shown by an example from digital assets, in which the difference is particularly pronounced.

The starting point is a portfolio invested entirely in Bitcoin. A Bitcoin position bought at the end of December 2021 stood 26.7% in the black at the end of June 2026; at the lowest month-end level in between, the investment was 64.2% in the red. The study examines what changes if half is shifted: into non-interest-bearing cash (the cash reference portfolio), into an interest-bearing money market reference (the SOFR reference portfolio), into one each of three defensive strategies, into a combination of these three in equal parts, referred to below as the combined building block, or, as a counter-check, into a basket of other cryptocurrencies. All portfolios are rebalanced once a year, in each case at year-end, to 50% each, the combined building block internally at the same date to one third each; in between, the weights move with prices. The period covers 54 months and thus essentially a single market cycle. The starting point lies a few weeks after Bitcoin’s then peak in November 2021 and thus at the beginning of a severe slump. The comparison with the starting portfolio of pure Bitcoin is therefore shaped substantially by this loss phase.

Four findings

1. The lower observed maximum drawdown comes almost entirely from the smaller Bitcoin position.

The maximum drawdown fell from 64.2% for pure Bitcoin to 32.1% with non-interest-bearing cash and 31.3% with the money market reference. The portfolio with the combined building block came to 30.6%, barely lower than the two reference portfolios. Of the individual strategies, only the stablecoin strategy lowered the maximum drawdown markedly further, to 25.5%. With the DeFi strategy it stood at 33.0%, with the market-neutral digital asset index at 33.5%, both times higher than with cash. In hindsight, the portfolio with stablecoin income alone would have performed better on performance and maximum drawdown than the portfolio with the combined building block. This selection, however, is not known in advance; the combined building block spreads the weight across three sources of return and thereby reduces dependence on any single one.

2. The higher performance cannot simply be read as diversification success.

With the combined building block, performance over the period was plus 91.4%, with the SOFR reference portfolio plus 69.1%. Without rebalancing, both cumulative returns would have been less than half as high: plus 23.3% for the SOFR reference portfolio, less than with pure Bitcoin, and plus 43.9% for the portfolio with the combined building block. The first rebalancing date, end of December 2022, fell on Bitcoin’s lowest month-end level, and rebalancing bought Bitcoin there. That is a property of this path and this date, not a general rule: depending on the calendar month of rebalancing, the portfolio with the combined building block would have reached between plus 56.9% and plus 91.4%.

The gap between the portfolio with the combined building block and the SOFR reference portfolio is 22.2 percentage points and reflects the higher historical return level of the defensive strategies. It reacts strongly to the return level of the two selected income strategies: if their excess return over the money market is arithmetically set to zero, its annual return advantage over the SOFR reference portfolio shrinks by 78%. The remainder is attributable almost entirely to the market-neutral digital asset index.

3. Under FTX stress, two differently constructed building blocks lost at the same time.

In the month of the FTX collapse, the DeFi strategy lost 4.8% and the market-neutral digital asset index 4.6%, in each case a multiple of their usual monthly fluctuation. Only the stablecoin strategy was in positive territory. Several strategies therefore do not automatically mean several independent risks. Whether a common risk factor lies behind the joint losses can, however, be neither demonstrated nor refuted with three crises in 54 months.

4. Smooth return series and monthly figures show less risk than there may be.

The stablecoin strategy was in positive territory in all 54 months. Its monthly returns are, moreover, strongly related to those of the previous month. Such a pattern can occur with smoothed valuations, which show lower fluctuation than unsmoothed ones. On its own, however, it does not prove smoothing. Its return premium over the money market can be understood economically as compensation for additional risks, such as counterparty, credit, custody and protocol risks. A calm return history therefore does not prove that these risks are low, only that they did not show up in the monthly returns of the period examined. Added to this is data frequency, because monthly figures can only underestimate declines. For the altcoin index, for which daily values are available, the price stood more than five percent below the previous month-end at some point within the month in 36 of 54 months, and ten of these months nevertheless closed in positive territory. For the three defensive strategies, the data used for this study contain no daily values.

What this means for your practice

Three questions follow from the findings that apply to any asset class, to the hedge fund in an equity portfolio just as to the absolute return fund or an allocation to private debt.

1. Against which realistic alternative is the new strategy compared?

If a provider shows that its strategy would have limited a portfolio’s loss markedly, it is worth asking in return what the same shift into the money market would have achieved. Also ask whether, how often and on which date rebalancing took place. For the portfolio of 50% Bitcoin and 50% combined building block, cumulative performance with annual rebalancing at year-end was 91.4%, without rebalancing 43.9%. This difference of 47.5 percentage points was more than twice as large as the lead over the SOFR reference portfolio of 22.2 percentage points. Depending on the rebalancing month chosen, cumulative performance ranged between 56.9% and 91.4%.

2. What happens exactly when the existing portfolio loses?

Average correlations and volatilities over long periods say little about whether building blocks fall together in a crisis. Have a strategy’s results shown for the worst months of your existing portfolio, not just its averages.

3. What does the historical lead depend on?

If a strategy appears both high-yielding and calm, two questions to the provider are worthwhile. Does the track record reflect actual investment results or a backtest? And what risk does the strategy bear for its excess return over the money market? In addition there is a calculation of your own: reduce the strategy’s monthly returns uniformly so far that over the period it would have earned only as much as a money market investment, and leave its fluctuation unchanged. If little remains of the return advantage, it depends substantially on the strategy’s high return level. The higher return alone then does not prove that it diversifies the portfolio better than a money market investment does.

The figures describe a completed period with selected individual strategies. They are neither a forecast nor an allocation recommendation. The questions, by contrast, can be put to any addition. Some comparisons in Part A were not part of the study design fixed in advance and were added subsequently: the money market reference, the calculations without rebalancing and with other rebalancing dates, and the threshold of the daily evaluation. Part B marks them individually. Part B contains the full calculation with study design, all metrics, the event test, the sensitivity to the return level, costs and implementability, data frequency and three methodological deep dives.

PART B · METHODOLOGY AND EVIDENCE

Research question and link to the previous study

The previous study, “Correlation, covariance and event risk”, the technical article on the Portfoliothese newsletter “What a correlation figure doesn’t tell you”, examined how a single addition changes the volatility of a Bitcoin portfolio. Three building blocks with correlations to Bitcoin between −0.08 and +0.59 led, with a 50% addition, to portfolio volatilities between 27.0 and 28.4%, and a plain cash allocation came to 27.1%. Most of the risk reduction came from replacing a highly volatile building block with a low-volatility one.

Two questions remained open. How do the same additions affect return and loss path? And what happens when several supposedly diversifying sources of return are combined? The occasion for the second question was the FTX month, in which the market-neutral digital asset index lost 4.6% and the DeFi income strategy 4.8%, in each case a multiple of their usual monthly fluctuation.

The main question of this study is therefore: does the combination of several defensive sources of return change the return and loss path of a digital asset portfolio differently from each of these sources individually, and do several strategies also carry several risk factors? Three working hypotheses structure the evaluation. H1 expects that the portfolio with the combined building block hardly differs from the individual additions at the volatility level. This expectation is largely predetermined arithmetically and is therefore tested in only one sentence. H2 leaves open whether the combined building block smooths the loss path of the defensive part. H3 states that several differently constructed building blocks can be hit by the same source of risk in certain events.

What was fixed in advance

Research question, building blocks, portfolios, metrics, events and evaluation rules were set out in a design protocol and frozen on 10 September 2026, before a single figure was calculated that was not already known from the previous study. Before the freeze the protocol went through three revised versions. None of them shifted a threshold or evaluation rule in a more favourable direction.

Every evaluation not contained in the protocol is post hoc and is marked in the text with an asterisk or explicitly. This concerns the comparison with a 50% SOFR reference, the figures on the highest month-end level, the counter-check of the combined building block without the stablecoin contribution, the thresholds of the daily evaluation, the observation on March 2023, the buy-and-hold comparison without rebalancing, the calculations with other rebalancing dates and the cost break-even. All deviations are kept in a dated change log; details on the procedure are in the section On the procedure.

The rebalancing convention is not among the points fixed in advance. The calculation uses annual rebalancing at year-end. Because the first rebalancing date fell on Bitcoin’s lowest month-end level, year-end yields the highest total returns of all twelve possible dates for the portfolio with the combined building block and both reference portfolios; the section Rebalancing and its date shows what this means for the results.

Data, portfolios and references

The same five building blocks are examined as in the previous study: Bitcoin via the Galaxy Bitcoin Series, the altcoin index via the S&P Cryptocurrency BDA Ex-MegaCap Index, the market-neutral digital asset index via the Galaxy VisionTrack Market Neutral Index and two anonymised individual strategies for stablecoin income and DeFi income. The window covers January 2022 to June 2026, that is 54 monthly returns. All returns are calculated in US dollars. For a euro investor without currency hedging, exchange-rate effects would be added. They change all end values by the same factor and leave their ranking unchanged, but can shift the loss paths and thus the maximum drawdown in euros.

The two income series are not segment benchmarks. They were selected for their suitability to represent the respective return driver, not for the level of their returns. Selection bias nevertheless cannot be ruled out. In addition there is a survivorship effect: only strategies that survived the entire period came into question. Because this study shows their returns prominently for the first time, it does not stop at a caveat: the section Sensitivity to the return level calculates how sensitively the results react to a lower return level. Readers cannot check the justification for the selection itself because of the anonymisation. The market-neutral digital asset index is based on returns that participating managers contribute voluntarily and that are not independently verified; reporting and late-reporting effects, so-called backfill, are known for such indices.

The combined building block consists of one third each of stablecoin income, DeFi income and the market-neutral digital asset index. The central study portfolio consists of 50% Bitcoin and 50% combined building block; it is referred to below as the portfolio with the combined building block. Alongside it are the 50% individual additions from the previous study. All portfolios and the combined building block itself are rebalanced once a year at year-end to their initial weights, in the counter-calculation correspondingly on 30 June; between the rebalancing dates the weights move with prices. The weights are comparison measures and not an allocation recommendation.

Two references put the results in context. Non-interest-bearing cash with zero percent return and fluctuation is the mechanical reference from the previous study. The SOFR reference is the compounded index of the secured US overnight rate of the Federal Reserve Bank of New York; its monthly return is the index level at month-end divided by that at the end of the previous month, minus one. It stands for an idealised money market investment without costs. The portfolio of 50% Bitcoin and 50% SOFR reference, referred to below as the SOFR reference portfolio, is post hoc. The portfolio of 50% Bitcoin and 50% cash is called the cash reference portfolio; together they are the two reference portfolios.

Metrics for all series

Over a window of this length, the observed maximum drawdown is essentially a single realised episode. It therefore describes the observed path and not a property of a building block, and it is always shown here together with the worst month and the recovery period.

For H1 one sentence suffices: the portfolio with the combined building block fluctuated at 28.0% a year and thus lies between the defensive individual additions at 27.4 to 28.7% and close to the 50% cash portfolio at 28.1%; the expectation is confirmed and carries no finding of its own.

The comparison benchmark

Compared with pure Bitcoin, the portfolio with the combined building block stands at plus 91.4% rather than plus 26.7% performance and a maximum drawdown 33.6 percentage points lower. How much of this is attributable to the combined building block depends on the comparison benchmark. The cash reference portfolio already reaches −32.1%, the SOFR reference portfolio −31.3%. For the maximum drawdown, the improvement thus arises predominantly from the smaller Bitcoin position; the portfolio with the combined building block lies only just below both reference portfolios.

On performance the same holds, albeit less markedly. The cash reference portfolio brings 29.9 percentage points over pure Bitcoin, the SOFR reference portfolio 42.4 percentage points. The remaining 22.2 percentage points separate the SOFR reference portfolio from the portfolio with the combined building block. These are comparison differences between defined portfolios and not a decomposition into causes; why the loss path of a portfolio cannot simply be assembled from its components is explained in the deep dive: Why the portfolio drawdown is not a weighted average of the individual drawdowns.

The two differ on recovery. The SOFR reference portfolio and the portfolio with the combined building block reached the low of their largest decline in December 2022; from there the SOFR reference portfolio needed ten months to regain the previous peak, the portfolio with the combined building block six. That peak, however, was reached at different points in time: for the SOFR reference portfolio at the starting value in December 2021, for the portfolio with the combined building block in March 2022.

The table shows a second decline that the maximum drawdown does not capture because it is smaller in percentage terms than the 2022 slump. From summer 2025, all portfolios with Bitcoin fell markedly: pure Bitcoin by 49.4% from the highest month-end level in July 2025, the portfolio with the combined building block by 24.5% from the highest month-end level in September 2025, the two reference portfolios by 27.8% and 26.2%. In this phase too, the defensive additions and the two reference portfolios differ by only a few percentage points. The figures on the highest month-end level are post hoc.

This article consistently reports returns and losses in percent and differences between portfolios in percentage points, in each case rounded to one decimal place. Differences are calculated from unrounded values and can therefore deviate in the last digit from the difference of the rounded figures. Dollar amounts depend on the initial capital and the level of wealth and are therefore unsuitable for comparison.

Rebalancing and its date

The comparison with buy and hold, that is, without rebalancing between Bitcoin and the second half, is post hoc. On this historical path, rebalancing had a considerable influence on performance. Without rebalancing the SOFR reference portfolio would have reached plus 23.3% instead of plus 69.1%, the portfolio with the combined building block plus 43.9% instead of plus 91.4%. Both reference portfolios would then lie below pure Bitcoin, the cash reference portfolio with plus 13.4%. For the maximum drawdown it changes little: for the portfolio with the combined building block it is the same in both cases at 30.6%, because the largest decline fell in the first year and thus before the first rebalancing date; for the SOFR reference portfolio it would be somewhat higher without rebalancing, at 32.7%. Rebalancing has hardly any effect on the gap between the two: 22.2 percentage points with and 20.6 percentage points without rebalancing.

Also post hoc is the question of how strongly the results depend on the rebalancing date. With each of the twelve calendar months as the rebalancing date, the performance of the portfolio with the combined building block ranges from plus 56.9% with an April date to plus 91.4% with a December date, and that of the SOFR reference portfolio from plus 37.2% to plus 69.1%. The cumulative returns thus depend heavily on a portfolio convention that has nothing to do with what drives the strategies’ returns. The comparisons between the portfolios hardly depend on it: its lead over the SOFR reference portfolio lies between 19.7 and 23.1 percentage points across all twelve dates, and its maximum drawdown at every date lies less than one percentage point from that of the SOFR reference portfolio. A general statement on the benefit of rebalancing does not follow from this one path.

The table compares year-end with 30 June and with no rebalancing. Without rebalancing, all portfolios with defensive sources of return are markedly lower; the maximum drawdown, by contrast, is identical for four of the seven portfolios because their largest decline fell in the first year and thus before the first rebalancing date. For the portfolio with the altcoin index, the path without rebalancing was even markedly better at minus 24.5%. 30 June as the rebalancing date lies close to the median of the twelve possible dates for all portfolios, whereas year-end is among the two most favourable for all portfolios except the one with altcoins. The stablecoin addition reacts most strongly. The strategy gained 16.4% in 2022 while Bitcoin lost 64.2%; at rebalancing at year-end, a good third of the stablecoin half was shifted into Bitcoin at the lowest month-end level, which rose by 156% in 2023 and 121% in 2024. The portfolio thus exceeds each of its two components on its own at plus 104.7%; without rebalancing it would have been plus 48.1%. One of the coming issues of the newsletter will be devoted in detail to the effects of rebalancing, including the choice of frequency and rebalancing date.

The combined building block

The combined building block comprises the three defensive sources of return stablecoin income, DeFi income and the market-neutral digital asset index. Their weights are reset to one third each on the respective annual rebalancing dates. On its own it achieved an annualised return of 11.2% with volatility of 2.7% and a maximum drawdown of 3.1%. Its maximum drawdown was thus below that of the DeFi income strategy at 6.9% and of the market-neutral digital asset index at 4.6%. From the low of this decline it regained the previous peak after three months. The stablecoin income strategy recorded no drawdown on the basis of month-end levels.

The worst month of the combined building block was November 2022 at −2.7%. Stablecoin income stood at +0.9% in that month, DeFi income at −4.8% and the market-neutral digital asset index at −4.6%; because the weights had shifted since the start of the year, the result is not an exact average of thirds. On H2 it can thus be stated: the combined building block ran more smoothly than two of its three components individually. In the FTX month, however, two of the three sources of return delivered clearly negative contributions at the same time, and only stablecoin income had a stabilising effect.

A post hoc counter-check replaces the stablecoin income strategy in the combined building block with non-interest-bearing cash and leaves the one-third weighting at the rebalancing dates unchanged. The maximum drawdown of this variant of the building block is 3.9% instead of 3.1%. The calculation compares two concretely defined building blocks. It does not decompose the drawdown into a weighting effect and a diversification effect, and it is to be read only in that way; the reasoning is in the deep dive: Why the portfolio drawdown is not a weighted average of the individual drawdowns.

The event test

For H3, the question of whether differently constructed building blocks are hit by the same source of risk in a crisis, a two-stage test had been defined.

Stage 1 tests whether several building blocks suffered conspicuously high losses in the same event month. For this, at least two differently constructed building blocks must each lose more than twice their own monthly volatility, that is, the usual fluctuation of their monthly returns over the entire study period.

Stage 2 then tests whether the losses of the affected building blocks fit the event’s previously recorded source of risk, for FTX for example exchange, counterparty and market-structure risks.

Only the five individual building blocks are tested. The combined building block is not counted in addition because it bundles sources of return that have already been examined. Its losses would therefore not be an independent additional observation.

Besides FTX, two further crises belong to the event test. In Terra/Luna, TerraUSD (UST) was an algorithmic stablecoin; in May 2022, UST, the associated LUNA token and large parts of the Terra ecosystem collapsed. In the USDC depeg of March 2023, the stablecoin USDC temporarily lost its dollar peg because part of its reserves was held at Silicon Valley Bank.

For each event it was recorded in advance which source of risk could burden the building blocks and which building blocks should therefore be affected. For Terra/Luna these were protocol and mechanism risks; strategies with exposure to algorithmic mechanisms or ecosystems built on them were expected to be affected, plus altcoins broadly. For FTX they were exchange, counterparty and market-structure risks. Strategies whose custody, trading or financing runs via central counterparties were expected to be affected. A strain going beyond the effect of the general market movement was expected. For the USDC depeg they were the reserve and bank risks of a stablecoin; strategies whose returns are tied to a dollar peg or to reserves at banks were expected to be affected. Because the depeg lasted only a few days, the interim strain can remain hidden in the month-end values.

Only for FTX did at least two differently constructed building blocks exceed the Stage 1 loss threshold. DeFi income and the market-neutral digital asset index each lost more than twice their monthly volatility at that time. For Terra/Luna, only the altcoin index exceeded this threshold; for the USDC depeg, no building block did.

It does not follow that the building blocks were exposed to different sources of risk in these two crises. The test merely found no joint strain there that met the defined threshold.

The FTX month was already known from the previous study. The loss threshold was set with knowledge of this case and aligned to it. That at least two building blocks exceeded the loss threshold in the FTX month is therefore not an independent confirmation of the test rule.

Stage 2 also remained open for FTX. It requires that the losses fit the previously recorded exchange, counterparty and market-structure risks. This cannot be established from the return series alone. That would require information on the positions held at the time and on the trading venues, custodians and financing partners used. For the strategies in the market-neutral index, such information is not sufficiently available. For the manager of the DeFi income strategy, it was not requested. Whether the losses fit the expected sources of risk therefore remains untested.

Under the rule fixed in advance, the result remains undecided. Beyond the already known FTX case, the test provides no additional confirmation of common risk factors. Nor, however, does it rule out such risk factors.

Only Terra/Luna and the USDC depeg were available for an independent confirmation. FTX must not be counted as a third event for this. This restriction had already been recorded in the protocol: in the most favourable case, a conspicuous joint strain would have shown up in both further events and could in each case have been attributed to the expected sources of risk.

Sensitivity to the return level

The calculation answers a practical question: how much of the historical return advantage of the portfolios with the two income strategies would remain if these strategies had earned less?

Both earned markedly more than the money market in the study period. The stablecoin income strategy achieved 12.4% a year against 4.1% for the SOFR reference, a premium of 8.3 percentage points. For the DeFi income strategy the premium was 5.6 percentage points. Both are selected individual strategies. Other strategies of the same type could earn less, and it is also open whether the strategies examined can hold their high premiums in future.

The study calculates two variants for this. In the first, the annualised premium of both income strategies over the SOFR reference is halved over the entire period; in the second it is zero. The combined building block and the portfolios with 50% Bitcoin are then recalculated.

For this, each monthly return of a strategy is lowered by a fixed amount. A separate deduction is determined for each strategy, which remains the same in all months. The fluctuation of the monthly returns and their temporal pattern, including their dependence on the previous month, remain unchanged.

The calculation thus tests dependence on the return level. It does not demonstrate a selection effect. Nor does it isolate the contribution of diversification or represent a stress scenario. Possible smoothing of the returns is not corrected, and the recalculated loss paths are also subject to the limitations of the original data.

Only the two selected income strategies are changed. The calculation was limited in advance to these two individual strategies. The market-neutral digital asset index, by contrast, bundles numerous strategies; its data and selection problems are not examined by this calculation. Its premium of 6.9 percentage points over the SOFR reference can be overstated by voluntary manager reporting, retrospectively added return data and survivorship. Its historical excess return remains included in both variants and thus supports the result of the combined building block. How sensitively the remaining return advantage reacts to deductions at the index is shown by the deep dive “How sensitively the lead reacts to deductions at the index”: after the premiums of the two income strategies have been neutralised, in the calculation used there a deduction at the index, stated as an annual rate, of 3.7% suffices to eliminate the return advantage of the portfolio with the combined building block over the SOFR reference portfolio.

The details are in the deep dive: The sensitivity calculation in detail.

The annual premium of the combined building block over the SOFR reference falls from 7.1 via 4.7 to 2.4 percentage points. Measured against the post hoc SOFR reference portfolio, the annual return advantage of the portfolio with the combined building block shrinks from 3.1 via 1.9 to 0.7 percentage points. It does not disappear entirely, but under this sensitivity falls by 78%. The remaining premium of the combined building block comes almost entirely from the market-neutral digital asset index, whose premium of 6.9 percentage points is not neutralised and enters at one third.

The loss path of the portfolio with the combined building block reacts little. The maximum drawdown moves from −30.6 via −31.6 to −32.5%, the worst month from −14.7 via −14.9 to −15.1%; recovery takes six months as observed, ten months in both variants.

The comparison with the individual additions fixed in advance yields the same pattern in all three variants: the portfolio with the combined building block lies between the individual additions on return and maximum drawdown. After neutralisation, the addition of the market-neutral digital asset index achieves the highest return at 14.8%, because its premium is not neutralised, and the addition of stablecoin income the lowest maximum drawdown at −29.0%. The portfolio with the combined building block achieves the highest return and the lowest maximum drawdown at the same time in no variant. In the observed series and at half premium, the stablecoin addition is superior to it on both measures; with neutralised premium this no longer holds for any individual addition. Whether the relationship between the two measures is favourable for it depends on the comparison benchmark and on investment objectives, and a retrospectively well-chosen individual addition presupposes a selection that is not known in advance. What is structurally given is the lower weight concentration on a single strategy. What has not been shown is that dependence on common risk factors falls as a result.

Costs and implementability

The returns are compared on different cost bases. For the two income strategies, fees have already been deducted. The index series, by contrast, contain no costs for access via a specific investment product. Compared with additions via an index, the income strategies therefore appear rather too unfavourable and the index rather too favourable.

This also affects the portfolio with the combined building block: measured by target weights, one sixth of the portfolio is in the market-neutral digital asset index. The costs of a corresponding investment would reduce its return contribution. How much this matters depends on the return of the income strategies; a computational stress test on this is contained in the deep dive “How sensitively the lead reacts to deductions at the index”. Implementation costs can also arise for Bitcoin and a money market investment.

Because the study does not compare specific investment products, the return advantage after all implementation costs cannot be determined. The results therefore do not show that the returns presented can actually be achieved with available products.

What monthly data do not show

Monthly values can only underestimate the maximum drawdown, because every month-end level is also a daily level. This applies to all series in this study, including Bitcoin.

For the two income strategies, the market-neutral digital asset index and the Bitcoin series used, only monthly values are available. For the altcoin index and the S&P Cryptocurrency Broad Digital Asset Index, a broad index for cryptocurrencies, daily closing levels exist; the daily series of the altcoin index reproduces all 55 month-end values exactly. The diagnosis fixed in advance compares the maximum drawdown on a daily and a monthly basis. For the altcoin index it is −81.4% on a daily basis and −79.3% on a monthly basis. The order of magnitude is thus robust, the course is not: on a daily basis the decline runs from 4 January 2022 to 11 September 2023, on a monthly basis from 31 December 2021 to 30 December 2022.

The monthly view with thresholds chosen post hoc shows what a month-end close conceals. In 36 of the 54 months, the altcoin index stood at some point within the month more than five percent below the previous month-end, in 25 months more than ten percent. Ten of the 36 months nevertheless closed positive, and of the 25 still three. The largest interim decline among the months that closed positive occurred in April 2025 at −12.8% with a month-end close of +5.6%.

Also post hoc is an observation on March 2023. The S&P Cryptocurrency Broad Digital Asset Index closed 14.4% below the February month-end on 9 March, in the sell-off before the closure of Silicon Valley Bank. The USDC depeg fell on the weekend of 11 and 12 March. The daily series used contains only weekday closing levels, and on Monday, 13 March, the index already stood 1.7% above the February month-end. The depeg thus lies between two data points. This limit belongs to the calendar coverage of the series used; daily data with weekend values could capture such an event.

It cannot be inferred from this how the smooth series behaved within a month. All that can be said is that the observation is missing precisely where abrupt risks would matter most.

Result and limitations

H1 is confirmed as an expectation. On H2, the combined building block ran more smoothly than two of its three components, but in the stress month two of them lost at the same time. H3 remains undecided. The comparison benchmark decides the interpretation: against pure Bitcoin the portfolio with the combined building block looks strong, against the two reference portfolios little remains on the maximum drawdown. On performance a lead remains that reacts markedly to the return level of two selected strategies: under the sensitivity, the annualised return advantage over the SOFR reference portfolio falls by 78%. The cumulative returns moreover depend heavily on the rebalancing date, the comparisons between the portfolios hardly at all.

The limitations are those of the data. The reported maximum drawdowns are predominantly shaped by the loss phase of 2022, and the decline from summer 2025 was not complete at the end of the window. Every drawdown is a single observation and not an estimate of a property. The income series are selected individual strategies, the market-neutral digital asset index is based on voluntary reporting, and monthly frequency conceals paths within the month. The calculations answer a question of risk and return over a completed period. They apply to the digital asset allocation on its own. Its effect on the overall portfolio depends on its weight and on the interplay with the other investments. They are not a model allocation and take into account neither implementability nor liquidity or access to the strategies, nor the transaction costs and taxes of rebalancing.

Deep dive: Why the portfolio drawdown is not a weighted average of the individual drawdowns

The variance of a portfolio is a quadratic form of weights, variances and covariances. It can be decomposed exactly into contributions, as the previous study showed. The maximum drawdown, by contrast, is a property of the path: the largest decline from a peak to a later trough. It depends on the order of the returns, on compounding, on rebalancing to the initial weights and on which peak a decline refers to.

An example shows the consequence. The 50% weighted sum of the maximum drawdowns of Bitcoin and the combined building block would give 33.7%. The actual maximum drawdown of the portfolio with the combined building block is 30.6%. Both components reach the low of their largest decline in December 2022, but their losses refer to different peaks: Bitcoin to December 2021, the combined building block to October 2022, the portfolio with the combined building block to March 2022. In addition, the weights shift with prices between the rebalancing dates. The drawdown of a portfolio must therefore be calculated from its joint wealth path. The example refutes simple averaging; it does not rule out every conceivable attribution of drawdown contributions.

This study therefore works with comparison differences between concretely defined portfolios. Each such difference answers a specific question, for example what changes if the second half is held interest-bearing rather than non-interest-bearing. Their sum does not yield a set of causes, and a different order of comparisons would produce different differences.

Deep dive: The sensitivity calculation in detail

A series’ premium is the difference between its geometrically annualised return and that of the SOFR reference over the same window. The annualised arithmetic mean of the monthly differences would be a different quantity. The two differ noticeably: for the stablecoin income strategy 8.3 against 7.8 percentage points, for the DeFi income strategy 5.6 against 5.4 percentage points.

For the main sensitivity, the same amount is deducted from each monthly return of an income series, determined numerically so that the geometrically annualised return of the shifted series equals the SOFR reference. For the stablecoin income strategy this is 0.65 percentage points per month, for the DeFi income strategy 0.44 percentage points. For the intermediate step, the deduction halves the premium; it is 0.32 and 0.22 percentage points per month.

A constant shift leaves the variance and autocorrelation of the monthly returns unchanged. The event months shift by the same amount. Holding them artificially fixed would be incompatible with preserving variance and autocorrelation. The event test therefore runs exclusively on the observed original series; sensitivity and event analysis are not overlaid.

Deep dive: How sensitively the lead reacts to deductions at the index

The following calculation was added post hoc. It tests how sensitively the return advantage of the portfolio with the combined building block over the SOFR reference portfolio reacts to additional deductions at the market-neutral component. For this, only its monthly return is lowered by a constant amount until the advantage disappears.

With the historical returns of the income strategies, this threshold, stated as an annual rate, is 17.9%. With the premium of the income strategies halved it is 10.5%, with the premium fully neutralised 3.7%. In each case, the figure given is twelve times the monthly deduction; it is not an annual return loss calculated with compound interest.

These values are computational thresholds and not estimates of actual product costs; actual fees can act differently from a constant deduction. They show: the lower the return premiums of the income strategies, the more sensitively the remaining advantage reacts to additional deductions at the market-neutral component, for example through the costs of access. Whether the index is investable at all, they do not answer.

On the procedure

The change log was not kept continuously but from 24 September 2026; the earlier entries are reconstructed. One calculation fixed in advance was carried out late: the comparison between the portfolio with the combined building block and the individual additions under the sensitivity was produced only after the newsletter text was finished. Its result does not change any statement of the newsletter. The buy-and-hold comparison was added post hoc following a reviewer’s comment.

Data basis, verification and notes

Bitcoin: Galaxy Bitcoin Series. Altcoin index: S&P Cryptocurrency BDA Ex-MegaCap Index. Market-neutral digital asset index: Galaxy VisionTrack Market Neutral Index. SOFR reference: SOFR index of the Federal Reserve Bank of New York. Daily data: S&P Cryptocurrency BDA Ex-MegaCap Index and S&P Cryptocurrency Broad Digital Asset Index, daily closing levels on weekdays. Stablecoin income and DeFi income: two anonymised individual strategies with monthly net histories after fees.

All metrics are calculated from the unrounded series and recalculated in a verification sheet separate from the analysis master. Rounding is applied only for presentation; a back-calculation from table values can deviate in the last digit.

---

Boris Jurczyk is the founder of 21Nexus Capital and is therefore economically connected to the topic of digital assets. 21Nexus Capital acts as adviser to a fund that invests, among other things, in stablecoin income, DeFi income and altcoin strategies. There are currently no business relationships with the two fund companies behind the anonymised income strategies; in particular, 21Nexus is not invested there.

This article is for information purposes only and constitutes neither investment advice nor a recommendation to buy or sell financial instruments or crypto assets. Crypto assets are subject to substantial price fluctuations; a total loss of the capital invested is possible. The figures shown refer to the period January 2022 to June 2026. Past performance is not a reliable indicator of future results.