It's more difficult to turn this intuition into an algorithm and quantify it. I suppose I need to come up with clever ways to localize a local bottoming or topping event and then characterize the shape of it.
What I am relatively good at is finding moments of extreme extension. Unless a major new trend is about to occur these can indicate reasonable entry points. For example, if the price is getting "far" away from a moving average and price action is respecting a boundary or throwing a good candlestick event then a robot can take action based on well defined situation.
There are also "warnings" available. If the price is moving very fast then what is considered a good entry point may be a lot riskier. This type of thing can be examined across multiple time frames to detect a larger scale event while trading on lower time periods.
The latest thought and robot experiment is to collect these little nuggets of interpretation. Accumulate the "value" of a list of reasons that it looks like a good time to trade and compare to the "value" of those reasons that a trade should not be entered.
To illustrate, let's assume our robot is "thinking" about entering a long trade on the USDJPY pair:
- Bid price is 8 pips below the m15 moving average (+ 0.08)
- Bid price is touching the lower h1 bollinger band (+ 0.10)
- Bid price is within one pip of middle h4 bollinger band (+ 0.05)
- The h1 bollinger band was recently pinched
- The previous m15 candle dropped 25 pips
Perhaps many robots would be better than people expected if they were able to look for more reasons not to enter a trade?