---
title: Do chart patterns predict crypto prices? We tested look-alikes
url: https://comparand.xyz/blog/do-chart-lookalikes-predict-crypto
description: In 3,600 tests on 12 crypto markets, chart look-alikes picked the next-day direction 50.22% of the time, below the 50.92% from always guessing up.
updated: '2026-10-21'
category: Data
---

# Do chart patterns predict crypto prices? We tested look-alikes

In 3,600 tests on 12 crypto markets, chart look-alikes picked the next-day direction 50.22% of the time, below the 50.92% from always guessing up.

Crypto chart look-alikes did not beat always answering “up” in our test. Across 3,600 tests on Binance and Coinbase using hourly charts from January 2020 through September 2026, our rule got the next 24-hour direction right 50.22% of the time. Always answering “up” on those same test dates would have been right 50.92% of the time.

We tested BTC, ETH, SOL, XRP, DOGE, ADA, AVAX, LINK, BNB and TRX on Binance, plus BTC and ETH on Coinbase. That makes twelve separate markets, with 300 test times each. After fees, its imagined next-day trades lost 0.07% on average. Binance prices came from perps, contracts that let traders take a side on price without an end date. Coinbase prices came from spot markets, where the coin itself trades.

Each search stayed within one coin on one exchange. BTC on Binance matched only earlier BTC on Binance. Markets that needed more history started testing later than January 2020.

For how often crypto rose with no pattern at all, see [Crypto base rates](https://comparand.xyz/blog/past-crypto-returns-not-forecasts).

## What exactly did we ask the past charts?

A chart look-alike, also called an analog, is an earlier stretch of prices with a similar shape. For each test, we used 48 hourly closing prices. Each is the last price recorded in a completed hour.

We found ten earlier matching stretches that did not overlap each other. Then we looked at what price had done after each match over the next 12, 24 and 48 hours.

To turn those cases into a rule we could score, we took their median return. A return is the percentage price change. To find the median, we sorted the results from lowest to highest and averaged the middle pair.

If that median was positive, the rule answered “up”; if negative, “down.” We compared that answer with what actually followed the test chart. This measured a specific rule. It cannot tell us whether every named pattern or chart method works.

*Diagram: Show a past matching chart and what followed it before the test time, with the test chart's future hidden until we check the answer.*
## Did look-alikes beat always answering “up”?

The table combines the twelve Binance and Coinbase markets listed above. Every row uses the same 3,600 test times, hourly prices and 48-close charts, from January 2020 through September 2026.

| Time after the test | Look-alike direction correct | Actual prices went up | Average rule return after fees |
| --- | ---: | ---: | ---: |
| 12 hours | 50.44% | 51.67% | -0.05% |
| 24 hours | 50.22% | 50.92% | -0.07% |
| 48 hours | 50.25% | 51.39% | -0.08% |

The rule got the direction right about as often as it got it wrong. It also scored lower than always answering “up” at every time span tested. We found no clear prediction advantage for this rule across the combined tests.

## Did it work for any single coin?

Not in a steady way. Twelve markets times three time spans gives 36 separate checks. The rule beat always answering “up” in 14 of them, tied in 2 and fell short in 20. Its average trade after fees was above zero in 13 of the 36.

With that many checks, a few will look good by luck alone. So one strong coin is weak evidence. For Bitcoin alone, see [what Bitcoin did after similar charts](https://comparand.xyz/blog/bitcoin-after-closest-past-charts).

A tip for your own research: when someone reports a pattern's hit rate, put the coin's up-rate on the same dates next to it. A rule has to beat that number, not 50%. Our 50.22% looks a bit better than a coin flip, but it was below the 50.92% from always answering “up”.

The “actual prices went up” column is a base rate: how often something happened without using a matching chart. It uses exactly the dates the rule was scored on, so a gap between the columns cannot come from using different dates.

We did not test whether chance alone could explain the small differences in scores. Markets can move together, so these tests are not all separate pieces of evidence.

## Why isn't a coin's normal up-rate always the same?

Consider Binance BTCUSDT, the Bitcoin perp priced in USDT, a token designed to track the dollar. In hourly price history from September 8, 2019, through October 20, 2026, we found 62,335 starting hours with a complete next day of prices. Price rose over that next day in 51.88% of those cases.

That share changed over time. For the same market and hourly prices, the next-day up-rate was 58.03% across 8,784 cases starting in 2020. It was 46.88% across 8,760 cases starting in 2022.

Those cases overlap: nearby starting hours share much of the same later price movement. So they are not that many separate events.

This is why the first table uses the same test dates for both direction columns. The longer Bitcoin history helps explain why dates matter, but it cannot replace a comparison on the same dates. Our guide to [why past crypto moves are not a forecast](https://comparand.xyz/blog/past-crypto-returns-not-forecasts) explains this further.

## How often did the closest cases disagree?

At the next-day mark, 62.72% of the same 3,600 hourly-chart tests had four through six rising matches out of ten. These tests cover the Binance and Coinbase markets listed above from January 2020 through September 2026. Similar-looking charts often had a nearly even split between rising and non-rising prices afterward.

The median hides that split. It can point upward while several of the closest cases fell. Looking at each past case shows disagreement that the middle result leaves out. Comparand’s shape_analogs search shows each past case and its later moves, so you can put the falling cases beside the rising ones. See [how often past crypto cases disagree](https://comparand.xyz/blog/when-past-crypto-cases-disagree) for a closer look.

## Did a strong year carry into the next one?

For Binance BTCUSDT, using the same hourly prices and 48-close matching rule, we got the next 48-hour direction right in 72.73% of 44 tests in 2023. In 2024, the score was 36.36% across another 44 tests.

Those small yearly groups do not tell us why the result changed. Showing only the stronger year would leave out how much worse the same rule did later.

When you read a claim about a pattern, check whether it shows every year, including the weak ones. A strong year alone does not show that the rule will keep working. Our test of [1,000 random trading rules](https://comparand.xyz/blog/overfitting-1000-random-trading-rules) shows how a rule that looks great on some dates can fail on later ones.

## What happened after trading fees?

We also measured imagined trades based on the median. A long position gains when prices rise; a short position gains when they fall. Across the same hourly-chart tests in the first table, the average next-day return was -0.07% after fees.

We charged 0.05% to enter and 0.05% to exit each imagined trade. We left out funding (payments between the two sides of a perp), the cost of borrowing coins for a spot short, the gap between buy and sell prices, getting a worse price than expected, and delays.

We scored each imagined trade separately, not as one growing account. The direction score also ignores move size: small gains from correct answers can be outweighed by larger losses from wrong ones.

## How we measured this

We used public hourly closing prices for the Binance and Coinbase markets and dates listed above, spacing tests roughly evenly through time. We put each chart's price changes on a common scale, compared shapes, and kept ten past matches that did not overlap each other. Every match and its later result were already known at the test time. We left out cases with missing prices, without choosing cases based on whether prices rose or fell. We scored the median's direction and imagined returns after entry and exit fees, leaving out the other costs described above. As the [methods guide](https://comparand.xyz/methods) explains, results can overlap or move together, and this test only covers selected coins that still trade today.

## Try this in Comparand

With Comparand connected to an AI helper, such as Claude or ChatGPT, ask: “Find the ten closest past Binance BTCUSDT charts to the last 48 completed hourly closing prices. Show each case and what happened 12, 24 and 48 hours later.” This uses the same chart length and later time spans as our test.

The chart search, called shape_analogs, returns:

- Past BTCUSDT charts on Binance at the same hourly spacing, one per past episode, with none overlapping your query chart.
- Each match’s price path and moves 12, 24 and 48 hours later, in basis points; 100 basis points equals 1%.
- The source dates behind the answer and what the call cost.

It does not calculate an average, median or hit rate (the share of cases that went a chosen way). Those summaries can hide disagreement. We calculated this article's results in our own test. Each shape_analogs search costs $0.01 from your prepaid credits.

You can [try 3 coins free](https://comparand.xyz/try). The [docs](https://comparand.xyz/docs) explain how to connect your AI helper and choose the search settings. These are past cases, not a forecast.

## The short version

- This look-alike rule did worse than always answering “up” in the combined test: 50.22% versus 50.92% for the next day.
- By market, it beat always answering “up” in only 14 of 36 checks.
- Compare results on the same dates, for the same coins and exchanges.
- Check cases that disagree and years when the rule did poorly.
- A direction score leaves out fees, other costs and the size of price moves.

## Questions people ask

**Do chart patterns predict crypto prices?**

Our rule got the next day's direction right in 50.22% of 3,600 hourly-chart tests on Binance and Coinbase from January 2020 through September 2026. Always answering up did better on those dates. This tested a specific rule, so it cannot settle whether every chart pattern works.

**What should chart matching be compared with?**

Compare it with how often the same coin on the same exchange rose over the same length of time. Use the same test dates, because the share of rising prices can change across years.

**Did following the closest charts make money after fees?**

No, on average. Across the same 3,600 Binance and Coinbase hourly-chart tests from January 2020 through September 2026, the next-day rule averaged -0.07% per imagined trade after fees. The test left out other costs.

**Does Comparand return a forecast probability?**

No. Its chart search shows each past match and how its price moved afterward. It does not calculate an average result or a share of successful cases.