Strategies · Library · Stocks
§3.17Hedges and structures
Machine learning – single-stock KNN
For one stock, describe each past day by features such as moving averages of price and volume. Find the past days most similar to today and average what happened next to forecast the coming days. Trade if the forecast clears a threshold.
- Asset class
- Stocks
- Held for
- Days
- Data
- ARCANE has part of the data
- Symphony
- Cannot be built as a symphony
The rule
- Signal
- Predicted return over the next few days from the nearest-neighbour days in the stock's own history.
- Entry and exit
- Go long above an upper threshold, short below a lower one; close when the forecast fades.
- Sizing
- One position per stock.
How it fails
- Overfitting to history
- Needs strict out-of-sample tests
- Number of neighbours is a free choice
Data it needs
- daily prices
- daily volume
Prices exist; volume is not part of ARCANE's listed data.
How ARCANE uses it
Alpha could describe it as a similarity forecast, but a sealed Call needs a tested model first.
Needs a trained nearest-neighbour model that the builder's blocks cannot express.
Source: Kakushadze and Serur (2018), 151 Trading Strategies, §3.17, p. 54 of the SSRN version. Described in our words. Hedges and structures is ARCANE’s filing, not the book’s.