Walters Naked Trading strategy is centered around a concept of investor fatigue and how to detect when the current trend is exhausted and is likely to reverse. I have not tested the strategy but I think it provides a nice framework for a well bounded ML problem that could be useful for the Apply Machine Learning for Investing meetup group.
I watched this presentation during my mandatory exercise time yesterday. It seems like a well bounded but non-trivial way get started testing various ML ideas. I added my ideas about how to think about automating for ML below.
Summarized from how Naked Trading Works (Reversal Trading) By Walter Peters on You Tube
Naked Trading Reversal Rules:
- Bar reverses direction (color) of Prior Bar
- Bar has Higher High and Lower Low than Prior Bar
- Top tail of reversal bar Must have room to left.
- Unique space is a place the price went recently but didn’t stay.
- Search backwards to find the first bar with intersecting price point. (Count # bars)
- Next candle must trade higher for Buys or Lower for Bears.
- Must be largest candle in last 6 (optimistic) 10 (conservative)
- Looking for largest candle we have seen in a long time.
- Set stop loss at the top of the Reversal Bar.
- Move stop loss to break even so nothing but commission is at risk when once in profit.
- Implied but not explicitly Stated
- It looked like he took a little more than reversal size as stop loss.
- Set the profit taker by moving the size of the reversal bar in same direction as the reversal bar.
- He mentioned staying in the trade until the system started showing signs of reversal but it looked like he was watch for shrinking bars but did not explicitly state the rule. Also did not state rule for moving stop loss or profit taker during this period.
Ideas for Automating Naked Trading Reversal using ML
He was testing on a 4 hour bar but there should be similar settings that would work on other time frames but the constants he used may need to be moved around. Here are some ideas for using ML techniques to find the variant that would work for different time frames.
- He is really describing a 3 bar entry decision but you can place the limit order based on end of the second trade provided it expires within 1 bar.
- His Largest bar constant in X bars is ideal candidate to find optimal value of X tested with Genetic algorithm.
- Minimum room or Ideal range of room to left is ideal candidate to be tested with genetic algorithm.
- Rule for next bar must take out low of reversal bar could be morphed to Y% of Next Z bars must have low below reversals bars and both Y and Z can be tested with a genetic algorithm. Could also measure the % by which next bar took exceeded the reversal bar.
- Does reversal bar really have to be larger than prior bar or is this really a symptom. Could this bar just be Y times larger than average of last Z bars. Seems like Ideal thing to test with a random optimizer.
Items Ideal for ML Tree, Bayes or KNN system:
- Amount of Room to Left before running into intersecting bar can be measured and used as feature.
- Direction and relative size of intersecting bar to left can be measured and used as features.
- Deepest valley for bull bars or Highest Peak for bear bars can be measured and used as Feature. Also the % of bars in the bottom 5% of this range can be measured and sued as a feature.
- Size reversal bar can be compared to Prior bars so Number of bars before encountering a equal or larger bar can be used as feature. Number of Bars since last reversal bar can be used as input to ML. ML Tree, Bayes or KNN system.
- % of bars moving with trend (same direction as prior bar) over last X bars can be used as input to ML.
- Whether this bar is moving in same direction as next longer trend can be used as a feature.
- Average Size of bars in last B% of Bars since reversals as compared to All bars since reversal could be used features to help determine Market Fatigue level for exit. Figuring out a Exit threshold and size of B may be better suited for genetic algorithm or random optimizer.