Automated algorithmic trading for crypto: what fees did to a year of BTC crossovers

Automated algorithmic trading for crypto means software that buys and sells coins on fixed rules, around the clock, through an exchange API. The rule matters less than how often it trades: on a year of hourly Bitcoin data, the classic 9/21 moving-average crossover traded 194 times and lost 24% before fees, then lost 49% once a 0.1% fee was charged on each buy and each sell.

AI answers to this question suggest a moving-average crossover as a sensible first strategy and mention fees in a footnote. So we ran three common versions of that rule on BTC and put the fee in the main table.

Three crossover rules, one year of BTC, three fee levels

Starting balance $1,000. Bitcoin itself fell 26.9% over the same period, so holding turned $1,000 into $731.

Before fees, all three rules lost less than holding, because they sat in cash for part of a falling year. After a 0.1% fee, the fastest rule lost almost twice as much as holding. The slowest rule still beat holding at 0.1%, and every rule still lost money.

Win rates were low for all three: 27%, 23% and 32% of round trips closed higher, and the median trade lost between 0.4% and 1%. Crossover rules live on a few long trends, so a run of small losses is normal for them, and each of those small losses also pays the fee twice.

How we counted: every hourly BTC-USD candle on Coinbase from 2 November 2025 to 24 October 2026 (after 200 hours of warm-up for the averages; 10 hours were missing from the feed). Long only, spot, no leverage, no stop. The rule buys at the close of the hour the fast average crosses above the slow one and sells at the close of the hour it crosses back. No slippage is included, so real fills would look slightly worse. The 0.1% and 0.6% rates are round examples; check your exchange's own fee page for your tier.

Hypothetical example: what 81 trades cost

• Rule: EMA 20 / EMA 50 on hourly BTC
• Balance: $1,000
• Fee: 0.1% on each buy and each sell
• Round trips in a year: 81, so 162 fills

Each fill costs about $1 on a $1,000 balance. Over the year the fees added up to about $130, and because each fee also shrank the balance the next trade used, the ending balance fell from $930 to $791. The rule needed about 13% more from its signals just to pay the exchange.

A rough check before you automate any rule: multiply its round trips per year by two fills, then by your fee. For the 9/21 rule that is 194 × 2 × 0.1% = 39% of the balance. If the backtest gain before fees is smaller than that number, the strategy pays the exchange and keeps the losses.

What to set before the bot goes live

1. Count trades per week in the backtest first. In the table above, trade count explains most of the gap between the three rules.
2. Use your real fee tier, maker or taker, in the backtest, plus some slippage.
3. Compare against buy and hold over the same dates. Losing less than the coin in a down year can still mean losing money.
4. Give the exchange API key trading permission only, with withdrawals turned off.
5. Run the rule on simulated money for a few weeks and check that live signals match the backtest.

The crypto technical analysis guide shows how often a common EMA and RSI rule fires on BTC, and the paper trading test guide shows how many trades you need before a result means much.

Watching a bot decide before it trades

In TradingWizard you pick a market and an asset, set up a trading bot, and watch what it does. Each bot shows what it read, why it traded or waited, and when it checks next. Demo trades use simulated money. No real orders are sent.

Set up a demo bot in /wiz and count how many hours it waits for each trade it takes.

These figures come from one coin over one falling year. A rising year would change the ending balances, but the fee arithmetic stays the same, and none of this is investment advice.