Digital Assets Report 2021 — Institutional Arrival

The 2021 digital asset cycle differed from the 2017 cycle described in the earlier report in one respect that changed most of its characteristics: institutions participated.

Corporate treasuries, hedge funds, some endowments and a number of traditional asset managers took positions. Regulated custody became available. Futures and options markets developed. Public companies held these assets on their balance sheets. Payment and infrastructure businesses in the sector listed publicly.

Several consequences followed from that single change, and they are more interesting than the price path.

The correlation profile changed. A substantial part of the institutional case for digital assets was diversification — the claim that these assets were uncorrelated with traditional markets and would therefore improve portfolio efficiency. Once institutions held them alongside other risk assets, and managed them within the same risk frameworks, they began behaving like other risk assets. The diversification argument was undermined by the act of acting on it.

Leverage accumulated invisibly. Lending platforms, derivatives and structured yield arrangements built leverage into the system. Much of this sat outside any reporting framework, so its scale was unknown until it unwound.

Yield-bearing arrangements introduced a partial anchor — assets that could be lent or staked for return did have a cash flow, which changes the valuation analysis from the 2017 report. But it introduced credit and counterparty risk that the underlying assets themselves do not carry, and the distinction between the two was frequently blurred.

Infrastructure improved genuinely, and that improvement is the cycle's most durable legacy — more durable than the price levels, which retraced substantially.

Why participation undermined the diversification case

This is the cycle's most instructive mechanism, and it is a general one that applies to any asset bought for diversification.

The original argument: digital assets had shown low correlation with equities and bonds. Adding an uncorrelated asset to a portfolio improves risk-adjusted return, even if the asset is individually volatile. This is standard portfolio theory and the reasoning was sound on the data available.

Why the correlation changed:

• Common holders. When the same investors hold digital assets and equities, and face the same pressures — risk limits, redemptions, margin calls — they transact in both simultaneously. Common ownership creates correlation regardless of what the assets are.
• Common risk frameworks. An institution managing digital assets within its overall risk budget reduces the position when total portfolio risk rises. Since portfolio risk rises when equities fall, digital assets get sold when equities fall.
• Common driver. As the 2020 and 2021 reports describe, the dominant market variable in this period was the discount rate and liquidity conditions. An asset held as a long-duration, high-risk allocation responds to that variable the same way other long-duration risk assets do.
• Leverage. Leveraged positions must be reduced when losses occur, and forced selling correlates with whatever caused the loss.

An asset is uncorrelated because different people hold it for different reasons. Once the same people hold it for the same reasons as everything else in their portfolio, the correlation follows. Diversification benefits erode as they are exploited.

This is not specific to digital assets. It applies to any asset purchased for diversification, and it is one of the more reliable disappointments in institutional investing: the diversifying property is real when measured, and diminishes as capital acts on the measurement. Commodities, hedge fund strategies and several alternative asset classes have followed the same path.

The practical implication is that a correlation measured historically is a statement about the past ownership structure of an asset. If the ownership structure changes — and inflows change it by definition — the correlation should be expected to change too.

Invisible leverage

Leverage accumulated in this cycle through channels that were largely outside any reporting framework, which is why its scale surprised almost everyone when it unwound.

The channels:

• Lending platforms that accepted digital assets as collateral and lent against them, allowing holders to obtain liquidity without selling — and to buy more.
• Derivatives offering leveraged exposure directly, including perpetual futures with high leverage available to retail participants.
• Yield arrangements where deposited assets were lent onward, sometimes through multiple layers, with the depositor unaware of the ultimate borrower.
• Cross-collateralisation in which one asset was pledged to borrow against, and the proceeds used to acquire another that was pledged again.

The critical feature is that none of this was systematically reported. In traditional markets, margin lending is reported, broker leverage is regulated and disclosed, and supervisors have visibility into aggregate positions. In this market, no equivalent existed. Individual platforms knew their own books; nobody knew the aggregate.

The consequence is a specific failure mode:

• Leverage builds during rising prices, because rising collateral values support more borrowing.
• A price decline triggers margin calls, forcing sales.
• Sales push prices lower, triggering further calls.
• Cross-collateralisation transmits stress between assets that appeared unrelated.
• Nobody can assess how far it will go, because the aggregate position is unknown.

This is the same feedback structure the 2018 report describes in the volatility unwind: a positioning event, where the size of the move is determined by what was owned rather than by any new information. The absence of reporting is what made it unforecastable — not the novelty of the assets.

Yield, and the anchor problem revisited

The 2017 report argues that digital assets lack cash flow and therefore lack a valuation anchor. 2021 complicated that in a way worth setting out carefully, because the complication was widely misunderstood.

Some digital assets generate yield. Assets can be staked to secure a network in exchange for rewards, or lent to borrowers for interest. These are genuine cash flows, and where they exist, discounted cash flow analysis becomes partially applicable. The 2017 analysis does not apply straightforwardly to a yielding asset.

But the yield's source determines what it actually is, and the distinction is critical:

• Protocol rewards — new units issued to those securing a network. This is a genuine cash flow, but it is paid in the same asset, and issuing new units dilutes existing holders. A yield paid in the asset itself is partly a redistribution among holders rather than a return from outside.
• Lending interest — a borrower pays to borrow the asset. This is a genuine external cash flow, but it carries the borrower's credit risk. The yield is compensation for the possibility of not being repaid.
• Fee revenue — a share of transaction fees paid by users. This is the closest analogue to a conventional cash flow: external, generated by economic activity, not dilutive.

The distinction between these was routinely obscured in how yields were presented. A headline rate combining protocol rewards, lending interest and fee revenue is not a single thing, and the risks attached to each component are entirely different.

The 2022 failures of several yield-generating arrangements resolved this. Depositors who believed they were earning a return on the asset discovered they had been unsecured creditors of an intermediary. The yield was compensation for counterparty risk, and the counterparty risk was realised.

A yield is a payment for bearing something. The analytical question is always what. In 2021 that question was frequently not asked, and in 2022 it was answered.

Infrastructure as the durable legacy

The most lasting outcome of the 2021 cycle is infrastructure, and it survived the price retracement.

Custody developed from a genuine operational obstacle into a solved problem. Regulated custodians, insured arrangements and institutional-grade key management became available. This mattered because custody had been a real barrier — fiduciaries cannot hold assets they cannot custody appropriately.

Regulated derivatives markets developed, providing hedging capability and price discovery in a supervised venue.

Accounting and audit practice developed, addressing how holdings are recognised and valued, which is a prerequisite for corporate and institutional holding.

Compliance tooling developed for transaction monitoring and sanctions screening, which is prerequisite for regulated institutions to participate at all.

These improvements did not reverse when prices fell. Infrastructure built during a cycle persists after it, which is the general pattern in emerging asset classes and is one reason each cycle differs structurally from the last even when the price path rhymes.

The 2017 cycle was retail and largely unintermediated. The 2021 cycle brought institutional infrastructure. Whatever follows will start from that infrastructure rather than from where 2017 started — which is the most substantive difference between the two cycles and the reason the comparison between them is limited.

The diversification paradox

The erosion of a diversification benefit through the act of exploiting it is worth developing as a general principle, because it applies to nearly every asset bought for its correlation properties and is consistently underestimated.

The setup. An asset shows low correlation with a portfolio's existing holdings. Portfolio theory says adding it improves risk-adjusted return. The measurement is real and the reasoning is sound.

Why acting on it changes it:

• The measured correlation is a property of who held the asset and why. Low correlation arises because the holders' reasons for buying and selling differ from the reasons that drive the existing portfolio. Change the holders and you change the correlation.
• Institutional adoption imports institutional behaviour. An institution holding the new asset inside its overall risk budget will reduce it when total portfolio risk rises — which happens when the existing portfolio falls. That single fact creates correlation regardless of what the asset is.
• Leverage synchronises it further. Leveraged positions must be reduced on losses, and forced selling correlates with whatever caused the loss.
• The flows themselves create a common factor. When the same allocation decisions drive purchases and sales across many institutions simultaneously, the asset's price responds to allocation flows rather than to its own fundamentals.

The consequence is that a diversification benefit is largest before it is widely exploited and smallest afterwards. An investor who allocates on the strength of a measured historical correlation is buying a property that their own purchase, aggregated across everyone reasoning similarly, erodes.

This is not unique to digital assets. Commodities, several hedge fund strategies, and multiple alternative asset classes have followed the same path: measured low correlation, institutional adoption, correlation rising, the diversification case weakening precisely as the allocation matured.

A historical correlation is a statement about the past ownership structure of an asset. Inflows change the ownership structure by definition, which means the correlation should be expected to change too — and the direction is predictable.

The practical response is not to avoid diversifiers but to treat measured correlation as decaying rather than fixed, to re-measure after adoption, and to be sceptical of an allocation case that rests entirely on a correlation observed during a period when the asset was held by a different population.

What an allocator could act on

Re-measure correlation after adoption, not once at allocation. The number that justified the allocation is likely to have changed by the time the allocation is material, and in a predictable direction.

Ask where the yield comes from. A headline rate combining protocol rewards, lending interest and fee revenue is not one thing. Protocol rewards paid in the same asset are partly a redistribution among holders. Lending interest carries the borrower's credit risk. Fee revenue is the only component resembling a conventional external cash flow. The 2022 failures resolved this: depositors who believed they were earning a return on an asset discovered they had been unsecured creditors of an intermediary.

Treat unreported leverage as unbounded leverage. In markets with no aggregate position reporting, the scale of leverage is unknowable and therefore the potential size of an unwind is unknowable. This is not a statement about the assets — it is a statement about the information environment, and it applies to any market without supervisory visibility.

Read the bankruptcy filings. The 2022 failures produced court records documenting exactly what depositors' claims were and where the yield originated. They are free and they are more informative than any contemporaneous marketing material.

Value the infrastructure separately from the prices. Custody, regulated derivatives, accounting practice and compliance tooling improved and persisted after prices retraced. Infrastructure built during a cycle survives it, which is why each cycle in an emerging asset class differs structurally from the last even when the price path rhymes.

What 2021 established for digital assets

• Diversification benefits erode as they are exploited, through common ownership, common risk frameworks and common drivers.
• Unreported leverage is unforecastable leverage, and produced a positioning event of the kind the 2018 report describes.
• Yield is compensation for bearing something, and the components of a headline yield carry entirely different risks.
• Infrastructure improvements proved durable where price levels did not.
• The core valuation problem is unchanged from 2017 for non-yielding assets — institutional participation altered who holds them, not how they are valued.