Strategies · Library · Crypto
§18.2Trend
Artificial neural network (ANN)
Train a neural network on technical indicators of Bitcoin returns, such as moving averages, volatility and RSI, to predict which return band comes next. Buy when the top band is predicted and sell when the bottom band is.
- Asset class
- Crypto
- Held for
- Intraday
- Data
- ARCANE has part of the data
- Symphony
- Cannot be built as a symphony
The rule
- Signal
- Predicted quantile of the next normalised return
- Entry and exit
- Trade each interval on the model's predicted band
- Sizing
- Single position, long or short
How it fails
- Overfitting
- Costs on frequent trading
- Model drift
Data it needs
- intraday BTC prices
- technical indicators
Daily BTC prices only; the method uses short intervals.
How ARCANE uses it
Alpha could treat a model output as one input to a Bitcoin Call, with caution. Paper only.
Needs intraday prices and a trained model.
Filed under the trend card
Hold each market in the direction of its own return over the past year, and size every position to the same expected volatility.
A paper plan from this card holds 21 to 252 days, stops at 2× the expected move and uses at most 100% of your risk budget. The card
Source: Kakushadze and Serur (2018), 151 Trading Strategies, §18.2, p. 116 of the SSRN version. Described in our words. Trend is ARCANE’s filing, not the book’s.