"Measure It, or Lose It"

Last week I argued your crypto portfolio isn't really diversified. Now the uncomfortable follow-up — here is how to prove it to yourself, and what to do about it.

Last week I told you that a ten-coin bag is one factor pretending to be many, and that position size matters more than breadth. Several readers wrote back with a fair objection: show me the work. Tell me how to actually measure this, and tell me how to act on it without drowning in spreadsheets. This piece does both. It is shorter on argument and longer on method. There are only two numbers a retail crypto investor genuinely needs to monitor — rolling correlation and position size as a fraction of survivable drawdown. Everything else is decoration. I will show you how to compute each, what the danger thresholds are, and why the most important decision in your portfolio is one you make before you buy any coin at all.

Number one: rolling correlation

Correlation is the single statistic that tells you whether your portfolio is what you think it is. The retail mistake is to look at it once, see a comfortable number, and assume the comfortable number is permanent. It is not. Crypto correlations are regime-dependent, and the regime shifts are precisely when correlation matters most. The only honest way to track correlation is to roll it.

A rolling correlation is Pearson's correlation coefficient calculated over a moving window of recent returns — typically 30 or 90 days for crypto. You recompute it daily. The window slides forward. What you end up with is not a single number but a time series: a chart of how tightly your assets have moved together, day by day, over the last year. That chart is the one that matters. If every line on it sits above 0.8 most of the time and spikes toward 0.95 during selloffs, your portfolio is one position. If some lines stay near zero or go negative during stress, you actually own something diversified.

You do not need a Bloomberg terminal. Python with pandas does this in six lines. A Google Sheets implementation with CORREL over a rolling range works. Glassnode, CoinMetrics, and Kaiko publish pairwise correlations for the majors directly. The tooling is trivial. What is not trivial is actually looking at the output. Most people refuse to, because the output tends to confirm something they paid real money to not know.

Correlation is a fair-weather friend. It stays low when you don't need the protection and climbs to one the moment you do.

Number two: position size as fraction of survivable drawdown

Now the second number, which is the one that actually decides whether you live to trade another cycle. Forget expected returns. Forget Sharpe ratios. Forget everything clever. The only question that matters for a retail crypto allocation is this: if the position falls 80 per cent tomorrow, does my total financial life still function? If yes, the size is defensible. If no, the size is wrong regardless of how confident you feel about the thesis.

This sounds trivial. It is not how anyone actually sizes positions. Most retail allocations are sized by a completely different logic, which is "how much do I feel like putting in". The result is that portfolios that look fine in calm markets reveal themselves in drawdowns as unsurvivable, and the person holding them is forced to sell at the bottom because their rent is due. The forced seller is always the person who sized the position as if drawdowns would not happen.

A cleaner version of the same discipline is the Kelly criterion, adapted. Full Kelly gives the position size that maximises long-run compound growth given a known edge — but full Kelly is too aggressive for any market where your probability estimates might be wrong, and in crypto they are always wrong. The industry standard is fractional Kelly: quarter-Kelly or half-Kelly. Half-Kelly captures roughly 75 per cent of the long-run growth rate of full Kelly while reducing drawdown variance by around 75 per cent. Quarter-Kelly is safer still and is what most professionals actually use in volatile regimes. In crypto, with regime shifts, leverage, and fat tails, quarter-Kelly or less is not conservative — it is honest.

The survivable-drawdown formula is the one I recommend to anyone who cannot tell you their historical win rate to three decimal places — which is everyone. It asks one question and answers it with arithmetic. What portfolio drawdown can I absorb without my life changing? Divide that by a realistic worst-case drawdown for the asset. The answer is your maximum allocation. Twelve per cent. Fifteen. Rarely more. Not because crypto is bad — but because an 80 per cent drawdown on a 40 per cent allocation wipes out a third of total net worth, and no thesis is worth that kind of forced error.

How the two numbers work together

Correlation tells you how many real bets you have. Position size tells you how large any single bet can be. They are not independent. If your rolling correlation shows that your ten coins are effectively one position, then your total crypto allocation — not the allocation to any single coin — is the position you must size against survivable drawdown. Most retail investors get this exactly backwards. They size each coin as if it were independent, treat the aggregate as diversified, and discover in the drawdown that the aggregate was the position all along. The correlation check prevents that error. The sizing check contains the damage when the correlation check fails.

A simple quarterly discipline captures both: once a quarter, pull 90 days of daily returns for every coin you hold. Compute rolling correlations to BTC. Plot them. If any coin has averaged above 0.85 for the quarter, consolidate it into BTC — you are paying basis risk for no additional factor. Then take your total crypto sleeve and divide it by your net worth. If the ratio exceeds your survivable-drawdown limit, trim it. Not tomorrow — today. The discipline is mechanical. The emotional cost of doing it is the only hard part, and the emotional cost of not doing it, when the next drawdown arrives, is considerably worse.

The takeaway

Two numbers. Rolling correlation, monitored. Position size, bounded by what you can survive. Neither requires a PhD, and neither gets discussed on the podcasts that sold you the ten-coin bag. That is not a coincidence. The discipline is boring, quiet, and financially decisive — which is the exact combination the content economy cannot monetise. Measure it yourself, or the market will measure it for you, and the market charges a considerably higher fee.