Bitcoin did well. Crypto as a basket did not.

Chosen by rule, with the dead coins kept in, the largest coins did worse than Bitcoin alone. Most of the universe collapses, the headline momentum effect did not show up, and the best portfolio rule still lost money.

01 · Baskets against Bitcoin

Without hindsight, the basket was the honest answer, and it did worse

The usual crypto backtest picks Bitcoin and Ether because they won. That guarantees a good result, and it is the survivorship trap. This study does something fairer. Every year-end it ranks all coins by trading volume, holds the largest 10 (or 50) for the next year, and keeps the coins that later died.

The data is built to be fair to the dead: 1,853 live and 6,404 dead coins, with the rules fixed before any return was seen. No stablecoins, wrapped or leveraged copies; at least a year of history and US$250,000 median daily volume. Prices are in Australian dollars. The balanced portfolio is 35% VAS (Australian shares), 30% VGS (global shares) and 35% VAF (bonds), rebalanced yearly.

From 2018 the top-10 basket made 9.4% a year against 23.5% for Bitcoin, with more volatility, and the top 50 lost 9.4% a year. The hindsight mix, 70% Bitcoin and 30% Ether, looks best, which is the point: it uses what we know now. The 2016 top-10 result also rests on a very thin universe, with only 3 coins eligible at end-2015 and 6 at end-2016.

02 · Most of the universe does not last

Being big in one year did not keep a coin alive

Of the 50 largest coins at end-2017, 44 had lost 90% or more within three years. The rate varied a lot by year, and the top 10 names were much sturdier, but the pattern is the same: a ranking is a snapshot, and the list changes.

The biggest coins by trading volume were not interchangeable either. Bitcoin was the largest every year, but the next four varied a great deal, and there was always a big winner somewhere that was not always the coin you would have picked.

Three-year return in Australian dollars, or to September 2026 where the window is shorter. Rankings are by 90-day median trading volume. After end-2017 XRP lost 91% and Ether made 1%; after end-2018 Ether made 2,608%. Holding all five equal-weight would have owned both the winners and the losers.

What about the names everyone knows now?

Seven large coins that trade from 2021, held alone and as a 5% slice of the balanced portfolio. All seven fell by two-thirds or more at some point.

Only Bitcoin clearly improved the portfolio's risk-adjusted return as a 5% slice (Sharpe 1.03 against 0.97 for the portfolio alone; Solana merely matched it at 0.97). A 5% Solana slice lifted the return to 45.5% a year but took the worst fall to −38%; Dogecoin cut the Sharpe ratio to 0.45. The coins that made the most gave the biggest falls, and the winners listed here are the ones that still exist.

03 · Do the famous factors work?

Momentum did not. Calm and no recent spikes did.

Liu, Tsyvinski and Wu (2022) documented market, size and momentum effects in crypto. We tested seven signals fixed in advance, run once on the largest 50 coins each year, one month ahead, with the dead coins included. Because seven tests were run, the bar is higher than the usual two standard errors: |t| of about 2.7.

Momentum, the headline factor in the academic papers, did not work here. Over one and three months it pointed the wrong way, and an independent recomputation of the three-month figure gave a t of −3.0. Size passed the bar but the wrong way round against the literature (volume may be standing in for liquidity). Two signals passed in both halves of the period: low volatility (+0.148, t +6.1) and avoiding coins with recent spikes (+0.120, t +5.7).

The lottery effect is mostly one bad fifth: coins that had just spiked lost 11% a year while the other four fifths made 31 to 45%. The same checks across different look-back windows gave the same answer: low volatility and lottery avoidance held at every window, and momentum was weakly negative over one to two months and nothing beyond.

Can you trade it?

A ranking effect is not a portfolio. We fixed two tests in advance and ran each once: take the 50 largest coins, drop the worst fifth on the signal, hold the rest equal-weight, rebalance monthly at 25 bps a trade, and compare with the plain equal-weight top 50. With two tests, t needs to reach about 2.2.

B clears the bar narrowly and A does not, so the low-volatility ranking effect did not turn into a portfolio result. Even B lost money: −2.7% a year over a period where Bitcoin made 23.5%. It is a better way to hold a bad basket, not a reason to hold the basket, and it is one test that passed narrowly on one cycle.

What this does not show

• Volume stands in for size. There are no market caps in the data, and volume can be wash-traded, so the size result may be about liquidity rather than size.
• One cycle. The sample is one bull, one bust and one recovery, dominated by a single market factor. One of the two ranking effects survived as a costed portfolio, and the sample is too short to be sure of even that.
• Thin early years. Only 3 coins were eligible at end-2015 and 6 at end-2016, so the 2016 numbers are close to a Bitcoin result by construction.
• Dead coins usually keep trading. Dead-coin exit rules (total loss, last price, or a rule) gave identical answers, because only 3 of 196 coins stopped trading while held. The model cannot capture being unable to sell.
• Filters were tightened after seeing member lists. Coin filters were tightened after inspecting member lists but before looking at any returns. Bad price prints were removed after they produced absurd volatility; real crashes were left alone.
• Tax, custody and exchange failure are outside it. Returns are before tax, and nothing here covers how the coins are held.