Crypto Drawdowns and the Math of Recovery

A price can finish a month above where it began and still suffer a meaningful decline along the way. Endpoint return answers where the journey ended relative to its start. Drawdown asks how far the price fell below a previous high during that journey.

The difference matters because a reader who entered near a local peak experienced a different path from a reader who measures only the beginning and end. Neither perspective becomes complete by adding more decimal places. The observation frequency, chosen window and treatment of missing prices shape the answer.

A historical sample, with its limits exposed

On 22 September 2026 at 19:31 UTC, CryptX retrieved CoinGecko’s 30-day Bitcoin chart through its connector. The response contained 120 ordered USD price observations, with endpoint timestamps spanning 23 August 2026, 20:00 UTC to 22 September 2026, 19:28:20 UTC.

The connector exposed those two endpoint timestamps, but did not supply a timestamp for each intermediate price. Our chart therefore labels intermediate points by observation number, not by invented dates or hours. The series is a historical sample; it is not a live quote panel or a full record of every trade.

Data provided by CoinGecko. Calculations are CryptX’s, based on the returned ordered observations. These figures describe this captured series, not Bitcoin’s lifetime drawdown or its condition when this article is published.

A running high supplies the reference point

Start with the first retained price as the initial running high. At each later observation, update that high if the new price exceeds it. Divide the new price by the running high, subtract one, and multiply by 100.

At a new high, the result is zero. Below a previous high, it is negative. The most negative value in the retained sequence is its maximum sampled drawdown. This method respects the order of events: a low that happened before a later high cannot be treated as a decline from that high.

In our sample, observation 45 supplied the relevant peak and observation 97 supplied the subsequent trough. The price recovered to at least that peak by observation 114. Because intermediate timestamps were absent, we can report those positions in the sequence but cannot responsibly assign exact peak, trough or recovery times.

There is also a boundary condition. Starting the calculation on 23 August ignores any peak before that retained window. An asset can be at a new high for this sample while remaining below a much earlier high. Window drawdown and decline from an all-time high answer different questions.

Why recovering a loss takes a bigger percentage

Imagine a hypothetical value falling from 100 to 80. That is a 20% loss, measured against 100. Returning from 80 to 100 requires an increase of 20 divided by 80: 25%. The base changed.

For a loss expressed as a positive fraction, the required recovery is the loss divided by one minus the loss. The relationship becomes steeper as the loss grows.

This is arithmetic, not an estimate of the likelihood or timing of recovery. Some assets never return to a previous price. A 100% loss leaves a zero base, so the formula has no finite recovery percentage.

The same logic explains our sampled Bitcoin figures: a decline of approximately 7.18% required an increase of approximately 7.73% from the trough to restore the earlier sampled peak. Equal-sized positive and negative percentage moves do not cancel.

Resolution changes what can be observed

A daily-close chart can miss an intraday low. A downsampled series can omit both a brief high and a brief low. Its maximum drawdown may therefore understate the decline visible in a richer, compatible record. It must not be presented as the exact worst tradable path.

CoinGecko’s historical chart documentation describes how the underlying API’s granularity depends on the requested range. The simplified connector we used returns a compact series rather than every raw timestamp-price pair. Those are separate layers of data handling, and the connector’s actual output sets the limit of what we can reproduce here.

Switching to exchange candles introduces another question. Bybit’s candle specification exposes open, high, low and close, but a candle alone does not reveal the complete order of intraday trades. A high and low within the same interval do not, by themselves, prove that the high happened first.

Price recovery is not account recovery

An asset’s price regaining a peak does not guarantee that a particular account has broken even. Fees, funding, withdrawals, deposits and position changes alter account outcomes. Leverage can also end a position before an eventual rebound; our unlevered price calculation models none of those events.

The SEC’s performance-claims guidance highlights the importance of understanding what costs and periods a performance presentation includes. Here the transparent description is narrower: a sampled USD price path, original arithmetic, and a clearly dated observation window.

A useful drawdown chart carries its reference high, observation frequency and recovery definition with it. Without those labels, a smooth line can hide the very experience the reader is trying to understand.

For a separate look at how a falling collateral price can affect a leveraged position, see DeFi health factors and liquidation.