I throughly enjoyed this chat with Stefan, especially the techniques he uses to identify the ideal markets for Mean Reversion trading. What was your #1 takeaway? Let me know what you think in the comments below.
It is really amazing, that i discovered mean reversion by chance and developed a trading strategy very similair to what Stefan describes, and it is the only strategy working for me!
Great interview. @33:20 it doesn't mean at all that FX pairs to the right are random, because Friedrichowski only did LL and SS, not LLL & SSS, LLLL & SSSS etc. Because many of his LL's might be a part of LLL or LLLL or LLLLL if you get my drift.
Great interview - recently starting getting into algotrading and this has got to be one of the more useful mean reversion interviews I've seen so far. Thanks for sharing!
You can make the bricks visually wide to communicate time. Because it could take a long time for a given brick to form and if you are a day trader, this is important.
This was a very interesting episode, but I walked away more confused than informed. I just started working on mean-reversion strategies about a month ago, thanks to your series with Cesar Alvarez. The thing I did not understand was the calculation behind the chart showing which forex pairs were better for mean reversion. I can see on my own chart how AUD/NZD work well for mean reversion, but I do not how to get to the point of determining. My own experience, matching Stefan's chart, is that EUR/USD is poor for mean reverting. I think I will need to re-watch/re-listen this episode several times over for it to make sense. In any case, thank you for all of the great podcasts and now videos!
To build that chart you have to do the following steps: 1) Create brick-candles (candles with fixed (e.g. percentage) size) for lets say the last ten years. To do that precisely, you need to start with tickdata (be aware that 10 years tickdata for 28 forex pairs are about 500GB data). 2) Count Long/Long, Short/Short, Long/Short and Short/Long combinations. 3) Combine LL with SS and SL with LS and you have exactly that chart. Since we see that for a couple of pairs we have an excess for reversals we got a good hint (argument) that trading reversals is a good idea for those pairs. If you would strictly use that observation as a complete trading set-up, you wait until a BRICK candle has finished and you would open a trade into to opposite direction (e.g. TP = brick_size - epsilon, SL = brick_size + epsilon). Be aware this extremely simple set-up might suffer to much from costs of trading, especially for smaller brick-sizes.. Therefore we need at least one additional filter => e.g. percentage price distance from an EMA.
Triangular equilibrium, order flow analysis and individual currency strength are very powerful tools to incorporate when developing Forex strategies. It's amazing what you get when you do not trade currency pairs so to speak, but you trade one currency (strongest) against the weakest. The challenge is timing, which is where order flow analysis comes in. It's never what you buy but when you buy it. Trading currencies is fascinating but very difficult because of triangular equilibrium.
@@JFDGroupLtd Thank you Stefan. I am building a mean-reversion forex system this week to feature on my site, and I was scanning all my old notes to find the chart you describe (time marker: ~25:50). You make it look magic, but it is really thoughtful analysis. Looking at NZD-CHF, it mean reverts well.
Great episode. I just have a little concern regarding the chart at 33:00, which shows weaker mean reversion factor in pairs like fiber and beast. I think the 0.08% change should be normalized to some appropriate measure of liquidity of pairs. One thing I noticed is that for the pairs on the left the avg spread is much closer to 0.08% as compared to avg spread for the pairs on the right. Something that enforces this issue is the fact that negative serial correlation (which is occurrence of SL or LS in this example) significantly coincides with high spread (illiquid) markets. Thanks.
Great and very valuable content! I like the brick explanation. I've been interested in the brick approach ever since I read the book of Jesse Livermore: "how to trade stocks" wherein his "record" is very similar to the brick approach. It is very surprising for me that there's a statistical perspective on bricks!
I've been trying to find any resource on how you can mathematically transform tick data to range/percentage bars. Does anyone have a link to this logic? or point me in the right direction?
The bricks are based on tick-data. You start at the very beginning. Let's say that the first tick has a price of 1.1 and you want to create brick-candles of 0.5%. Then you wait for next tick which is either greater than 1.1055 (1.1+0.5%) or lower then 1.0945 (1.1-0.5%). If that happens then the brick is finished and a new brick is opened.
@@JFDGroupLtd Is there any major risk in using closing price of Minute Data instead? I assume you would lose details around the absolute tops and bottoms printed by the high/low of minute bars. But would this impact the analysis you've conducted?
@@karunkrishna1111 It depends on the percentage brick size, which you have in mind. If you want to create bricks of 0.1%, your M1 data are fair enough to create those bricks. But if you want to create bricks with 0.02%, M1 data are not detailed enough. But you might do your own cross-check: if you want to create 0.1% bricks, then you might check, how many M1 candles are bigger (high-low) percentage wise than 0.1%. The lower that number the closer you are at the "real" bricks".
The bricks are based on tick-data. You start at the very beginning. Let's say that the first tick has a price of 1.1 and you want to create brick-candles of 0.5%. Then you wait for next tick which is either greater than 1.1055 (1.1+0.5%) or lower then 1.0945 (1.1-0.5%). If that happens then the brick is finished and a new brick is opened.
