DislikedI use the same approach also with great success since long time. Anyway I found out, that sometimes on every market long and strong trends happen, which could generate a higher than usual draw down. I think about to find a way to implement a trend following strategy which kicks in, when such a strong trend occurs and which relative's the draw down a bit… any ideas how to do this? i tested also some hedging approaches, but they work semi good in my opinion, because you never know when to stop the hedge…Ignored
The kurtosis on this distribution is crazy high, making it so that for a lot of pairs, the distribution should have these characteristics or something close/similar:
- 96% of candles will close within 1 SD. In a normal distribution 68.2% of candles will.
- 97% of candles will close within 2 SD. In a normal distribution 95% of candles will.
- 98% of candles will close within 3 SD. In a normal distribution 99.8% of candles will.
Check this part of the description of their Z6+ system:
"The red line is the maximum price movement in pips over the period given on the X axis in hours. The black line is the same maximum movement if the markets were perfectly efficient, simulated by randomizing the price curve."
They call their system an adaptive grid, but that is just dollar cost averaging with more levels, right? ;-)
- #376
- Edited 7:00am Nov 29, 2025 6:49am | Edited 7:00am
Inspired to look at this again, I calculated the price movement between two adjacent M1 bars (open-open prices) throughout the day. Note the difference between average and median. And what happens at 15:30 and 17:00 hours?! More to investigate...
- #377
- Edited 1:02pm Nov 29, 2025 8:31am | Edited 1:02pm
I tried to reproduce the graph made by the Zorro guys (see post #375). I generated 10,000 fake AUDUSD charts by for each M1 bar selecting a price movements at that same time of day at random from the real AUDUSD chart. This to ensure we get the same average volatility throughout the day as the real AUDUSD as shown in the graph of my previous post above.
Next I calculated the drift from the price at 01:00 in the morning. This graph shows the median price drift of the real and the synthetic AUDUSD charts:
So what I get is actually the opposite of the graph of the Zorro guys: the real price action is less strong than what you expect from a synthetic random walk?!
Edit: I just noticed they plot the "MaxDev", not the average. Ok, let's try that! Here is a graph showing the maximum* price movement:
*) actually the 95th-percentile to filter out the most idiotic extremes
Now both the real and synthetic charts show similar price drift.
I am confused. What have these Zorro guys plotted in their graph?
Next I calculated the drift from the price at 01:00 in the morning. This graph shows the median price drift of the real and the synthetic AUDUSD charts:
So what I get is actually the opposite of the graph of the Zorro guys: the real price action is less strong than what you expect from a synthetic random walk?!
Edit: I just noticed they plot the "MaxDev", not the average. Ok, let's try that! Here is a graph showing the maximum* price movement:
*) actually the 95th-percentile to filter out the most idiotic extremes
Now both the real and synthetic charts show similar price drift.
I am confused. What have these Zorro guys plotted in their graph?
- #378
- Edited Nov 30, 2025 7:42am Nov 29, 2025 11:56pm | Edited Nov 30, 2025 7:42am
I tried to reproduce the graph made by the Zorro guys (see post #375). I generated 10,000 fake AUDUSD charts by for each M1 bar selecting a price movements at that same time of day at random from the real AUDUSD chart. This to ensure we get the same average volatility throughout the day as the real AUDUSD as shown in the graph of my previous post above. Next I calculated the drift from the price at 01:00 in the morning. This graph shows the median price drift of the real...
Thanks for sharing the zorro stuff. I checked it out and doing a quick spot check on AUDUSD my graph looks similiar to theirs:
The blue line is the actual price, and the red line is a GBM generated price.
They almost surely are just measuring the max range (high - low) for each period and comparing them. You graph looked different because you took only the 95th percentile.
In a normal market this would make sense to do, but because the biggest deviation from normal prices have in terms of characteristics, is their kurtosis, you almost want to do the reverse and only look at the 5th percentile.
Take a second look at the SD probabilities I posted earlier.
- 96% of candles will close within 1 SD. In a normal distribution 68.2% of candles will.
- 97% of candles will close within 2 SD. In a normal distribution 95% of candles will.
- 98% of candles will close within 3 SD. In a normal distribution 99.8% of candles will.
These are applying to MA candles, which have much larger kurtosis than normal OHLC candles but the idea is the same. If anything, its more important because their is less kurtosis to take advantage of in OHLC candles. But the main point is, when there is high kurtosis, more occurrences fall within 2SD than normal BUT less occurrences fall within 3SD, like a lot less. Looking at these numbers you can see that the "action" in terms of opportunities or risk from kurtosis lies outside 3SD or inside 2 SD,
By trimming the 95th percentile, you took out all the occurrences that create the kurtosis. Any excess kurtosis will push more occurrences into the tail, which will be trimmed, and the higher the kurtosis the even more occurrences will be pushed to the tail. Even though the tail is MASSIVE MA candles, when looking at them we need to remember 3SD in a normal distribution only accounts for 0.2% of a normal distribution, so even though these 3SD events could occur 10-20x more than they should (in numbers I posted they are occurring 10x more than normal), trimming the 95th percentile would remove all of them under the guise of "noise".
The kurtosis is also why in your other graph the average is always higher than the median (this is inherent in any distribution that has excess kurtosis and is essentially measuring it, the large the distance, the high the excess kurtosis).
ri · skuht
