[pcabc]

Searching for warnings of the recent drop programatically

I've been looking at my breadth charts to see if there have been any hints that could be picked up on programmatically in the time preceding the recent drop.  The chart is really busy, too busy perhaps, as I've added SMAs to various curves and also indicator bars below each curve so I can see the status of the logic on whether there is potential danger.  Really this is work in progress, but the takeout for anyone who is interested is which parameters seem to be more useful as that may give pointers for anyone else digging into this.

There is a real danger here of trying to create a meta indicator that is so tuned for recent conditions that it is only relevent for this specific set of circumstances.

   

The chart has been annotated with black numbers.

1. This is my aggregate breadth position with black being DANGER.  On this timescale it looks like I could have given myself two weeks warning of the market move.  However, on a longer timescale there are things that might be false alarms.  This needs more work, it is not a magic bullet.

2. Comparison of % stocks above their 200 day EMAs, when compared to the 20 and 80 day SMA of this curve looks to be useful.

3. As above, but for the % stocks above their 50 day EMAs looks to be less useful as it gives a lot of flase alarms used on its own.  Here I am using the postion of the 10 day SMA to the 20 day SMA and the slope of the 20 day SMA as a warning.

4. New highs minus new lows does not appear to provide much of use.  I've averaged the new highs minus new lows to give an idea of direction, but it only confirms what is already happening.  The same appears true when comparing cumulative new highs minus new lows against its averages.

5.  I've not used the advance / decline line yet.  Spotting divergences is hard.  Comparing the A/D line against its moving averages only appears to confirm a move that is already happening.  Comparing it against a faster average such as a 20 day will create a lot of noise.  Perhaps I can do something with channels?  Or perhaps the relative strength between price and A/D?  Needs work.

6. Advancing / declining volume, comments against 5 apply.