This project is not maintained anymore.

After many years of working on Gekko, I’ve decided to stop my involvement in maintaining this project. You can read more about this decision on medium.

I’m now putting all my focus on my new prop trading firm Folkvang. You can find an article about that here on Coindesk.

If you’re interested in following this new journey, feel free to add me on Twitter.

Best of luck to everyone in their trading. So long, and thanks for all the fish!

Fork me on GitHub

Backtesting #

Gekko supports backtesting strategies over historical data. A Backtest is a simulation where you simulate running a strategy over a long time (such as the last 30 days) in a matter of seconds. Backtesting requires having market data locally available already. After a backtest Gekko will provide statistics about the market and the strategy's performance.

screen shot of gekko backtesting

Important things to remember:

Simplified simulation #

Simulating trades is done through a module called the paper trader. This module will use market candles together with fee, slippage and spread numbers to estimate trade executions costs. While the default settings work great for most big markets (USD/BTC or BTC/ETH), it becomes a lot less acurate on smaller markets with low volume and liquidity.

In live trading the notion of the "price" is more complicated than a single number. Both spread and slippage will effect your trade prices: these numbers describe your desired trades in relation to what people are currently offering in the market (this is called the orderbook). Read more about this in this explanation.

If you look at the following backtest result:

screen shot of backtesting at an illiquid market

You can see a lot of "spikes" of the price moving up and down. These are not actually price fluctuations but simply trades that happen on both sides of the orderbook (a bid is taken then an ask is taken). How far it jumps up and down is the spread (between the asks and the bids). In these cases the statistics from the simulation won't be very accurate (unless you configured a higher slippage to account for the spread). This is unfortunately a limitation in Gekko's backtesting model.