What each start date returned

What this means for a holder

The single most common way to be misled about this asset class is to be shown one number: an average return, a best case, or the outcome of one well-chosen start date. The honest version is the whole distribution, and the distribution here says two different things depending on how long you look.

Over one year, a bitcoin start date in our history ended below where it began 27.0% of the time, with results running from −83% to +9,051%. Over four years, none of them did, and the weakest still finished +31%. Ether tells a rougher version of the same story: 38.5% of one-year starts were losses, and 10.9% of four-year ones.

What that does not establish is a rule. Every four-year window in this data contains a bull market, because no four-year stretch of this market has yet failed to contain one — that is a statement about a nine-year sample, not about the asset. The drawdowns inside those winning windows were brutal: the same four-year period that ended +1,170% in the middle of the distribution contained falls of more than half along the way. A result you only collect if you hold through that is not the same thing as a result you are likely to collect.

The way we would read it: as a scale for your own patience, not as an expectation. If a one-year horizon is what you have, the honest planning number is the width of that spread, not its middle. If four years is genuinely available, this history says the entry date mattered much less than the holding did — in this history.

What this page cannot tell you

• The sample is small. Row counts in the thousands come from overlapping windows. At four years there are about 3 independent observations per asset. Any statement of the form “X% of the time” is a description of one price history.
• It starts in 18 Aug 2011. Fewer than two complete halving cycles, and none of the early history that competitors chart from 2011 — we do not hold verified closes for it, and we do not borrow them.
• Long horizons describe old conditions. The last scoreable four-year start is 27 Jul 2022. Nothing since then has finished.
• Survivorship is baked in. These two assets are here because they lasted. The same calculation over the coins that did not would look nothing like this, and it is not a calculation we can run.
• No fees, no taxes, no missed buys. Real results are lower by whatever those cost, which is not a rounding error over four years.

Frequently asked questions

What is a holding period return?

The return you would have got by buying once on a given day and selling exactly N days later, with nothing bought or sold in between. This page computes it for every single day in our price history and shows the whole spread of results, not an average.

Has holding bitcoin for four years ever lost money?

Not in our data, which starts on 18 Aug 2011. Of the 3,997 start dates with four years of history after them, none ended below where they began; the median gain was +1,170% and the weakest four-year stretch still finished +31%. That is a striking record and a very small one: those 3,997 windows overlap almost completely and amount to roughly 3 independent periods. Ether, over the same horizon, was below water on 11% of start dates.

Why do you show the minimum and the maximum instead of an average?

Because the average is the least useful number here. Over one year bitcoin start dates ran from −83% to +9,051%; a single mean drawn from that says almost nothing about what any particular year looked like. The spread is the finding. We show the median because it is robust, and the 10th and 90th percentiles so you can see how wide the middle really is.

Does this prove that holding longer is safer?

It shows that in this price history, longer holding periods had fewer loss-making start dates. It cannot prove the general claim, for three reasons stated plainly: our history begins 18 Aug 2011 for bitcoin and 9 Mar 2016 for ether, which is still a handful of cycles and not a law; the windows overlap, so the sample is far smaller than the row counts suggest; and every four-year window in it contains at least one enormous bull market, because there has not yet been a four-year stretch of this market that did not.

Why does the four-year table stop four years ago?

A four-year holding period needs four years of future to measure. The most recent start date we can score is 27 Jul 2022 — everything after it is still running. So the long horizons describe older market conditions than the short ones, and that is a bias towards the past rather than noise.

Are fees and taxes included?

No. These are raw daily closes: no exchange fee, no spread, no tax, and a purchase settled at the close rather than at the moment you would have pressed the button. The DCA calculator carries the same caveat with the typical fee ranges spelled out.

How this page is calculated

• Source. Daily closes in UTC from CoinGecko with Binance as a fallback, from 18 Aug 2011 to 26 Jul 2026. The same series that feeds the Hodlometer Index and the DCA calculator.
• One window per calendar day. For each day d we take the close on d and the close on d plus the horizon in calendar days, and record the change. Days without that much future are not scored rather than being cut short.
• Percentiles are linear-interpolated quantiles of the scored windows — the same helper used on the Mayer page and elsewhere, so the convention is identical across the site.
• “Ended lower” counts windows whose return is below zero, measured close to close. It is not a measure of how far underwater the position went in between: for that, see the drawdown episodes on Markets.
• No forecast is derived from any of this. The page publishes what happened at every entry point and stops there.

Data through 26 Jul 2026 · JSON API · All tools