Crypto Is Not Gold, Not Stocks, Not Anything: Liu and Tsyvinski Run the Numbers

Every new asset class arrives with a story attached. Bitcoin's stories were plentiful and contradictory. It was digital gold, a hedge against inflation and currency debasement. It was a tech stock, a levered bet on the adoption of a new technology. It was a commodity, whose value should track the cost of producing it, meaning electricity and mining hardware. It was a currency, whose value should respond to macroeconomic conditions.

These are not vibes. They are testable claims, and each one implies that crypto returns should be related to something specific: to gold prices, to equity factors, to energy costs, to macro variables.

Yukun Liu and Aleh Tsyvinski simply tested all of them.

The problem: nobody knew what kind of thing this was

If you are an investor deciding whether to hold cryptocurrency, the first question is not "will it go up." It is "what is it exposed to?" That determines whether it diversifies your portfolio or doubles down on risks you already have. An asset that behaves like a tech stock adds nothing to a portfolio of tech stocks. An asset that is genuinely unrelated to everything else is valuable even if its own returns are unimpressive.

The problem was that nobody had done the work. Crypto commentary was almost entirely narrative. What was missing was the boring, essential exercise: take the returns, regress them on the standard set of risk factors that finance has spent fifty years building, and see what sticks.

The key idea via analogy: testing every key on the ring

Imagine you have a locked door and a ring of keys, each labelled with a theory. You try each one.

Liu and Tsyvinski tried them all on Bitcoin, Ripple and Ethereum, the three largest coins at the time:

• Stock market factors. The market, size, value, momentum, profitability, investment, the whole standard toolkit that explains equity returns. No meaningful exposure.
• Macroeconomic factors. Industrial production, personal income, consumption, inflation, durable goods. If crypto is a currency or a macro hedge, these should matter. No meaningful exposure.
• Currency factors. Returns on major currencies against the dollar. No meaningful exposure.
• Commodity factors. Gold, precious metals, energy. This is the one the "digital gold" narrative most needs. No meaningful exposure.
• Cryptocurrency production factors. This is the clever one. If crypto is a commodity, its value should be tied to the cost of producing it. So they tested exposure to the cost of mining: electricity prices, and the stock returns of the companies that make mining hardware (the graphics card manufacturers). No meaningful exposure.

The key did not fit any lock. Cryptocurrency returns were essentially unexplained by every established risk factor in finance. Whatever this asset is, it is not a repackaging of something we already had.

Then they asked the second question: if the standard factors do not explain it, what does?

Momentum works, powerfully. Crypto exhibits strong time-series momentum: coins that have gone up recently keep going up, over horizons from a week to several weeks. This is not a subtle statistical effect, it is one of the strongest momentum signals documented in any asset class. A simple rule of buying after strong recent performance produced large returns in their sample.

Attention works. Proxies for investor attention, how much people are searching for and talking about crypto, strongly forecast future returns. When attention rises, prices follow. Importantly, this cuts both ways: they also find that negative attention, measured by things like spikes in searches for terms suggesting distress, forecasts lower returns.

Network factors matter. Measures of how many people are actually using the network, such as the number of wallet users and active addresses, are related to returns. That at least gestures at a fundamental: a payment network with more users is worth more, which is the Metcalfe-flavoured logic that Cong, Li and Wang later built into a full valuation model.

So the emerging picture of what drives crypto returns is: not risk, but attention and flow. Prices move because people notice, people buy, and other people notice that people are buying.

Why it mattered

• It settled the "digital gold" question empirically. Bitcoin's returns are not related to gold's returns, to inflation, or to any macro variable that a real inflation hedge would respond to. The narrative may still be true one day. In the data, it was not.
• It established crypto as a distinct asset class, for better and worse. The zero exposure to every known factor means crypto genuinely offered diversification. It also means finance's entire apparatus for understanding why an asset earns a return had nothing to say about it. An asset with no risk exposures has no risk premium, which raises the uncomfortable question of where the returns are supposed to come from at all.
• It made crypto a legitimate object of academic study. Doing the boring, rigorous factor work is what moves a topic from commentary to research. This paper, in a top journal, did that.
• It documented a real, tradeable anomaly. The strength of momentum and the predictive power of attention in crypto are among the largest such effects ever documented. Both have since become the backbone of systematic crypto strategies.

The honest limitations

• The sample is short and it is a bubble. The data cover a handful of years dominated by one of the most extreme speculative episodes in financial history. Statistical relationships estimated over that window are fragile. It is entirely possible that "no factor exposure" simply means "the sample was too short and too wild for any factor to show up."
• Absence of evidence, and all that. Failing to find a relationship with gold over a few years is not proof that none exists. In a longer, calmer sample, crypto might well develop the exposures the narratives claim.
• Momentum in a bubble is not a discovery. Any asset in a parabolic run will show enormous momentum. The finding is real, but attributing it to a durable behavioural effect rather than to the shape of a bubble is a genuine interpretive risk, and time-series momentum in crypto has been considerably less reliable since the sample ended.
• Attention predicting returns is uncomfortably circular. People search more when the price is rising, and the price rises when people buy. Disentangling "attention causes returns" from "attention and returns are both driven by the same underlying frenzy" is very hard, and the paper's careful work does not fully escape it.
• Nothing here says crypto is worth anything. The paper studies returns and exposures. It is explicitly not a valuation. An asset can have strong momentum, high attention sensitivity, no factor exposure, and a fundamental value of zero.

The one-line takeaway

Liu and Tsyvinski tested cryptocurrency against every risk factor finance has, found that it is exposed to none of them, is not digital gold, and is not a tech stock, and that what actually predicts crypto returns is momentum and sheer investor attention, which is a portrait of an asset driven by flow rather than fundamentals.