Similarity Thoughts
Dear Peter,
I hope your are fine. I wrote this page to summarize my thoughts about the theories and things you have presented. I have to admit that your way of thinking and approaching the fx world is pretty unique. However, I was never able to make it work. The reason is that there seems to be always (not just in your approaches presented) a trade-off between the probability of winning and the potential losses, i.e. the higher my winning probability becomes, the higher are the potential losses, too.
So, my question is not really directly related to your strategies (I know that you won't talk about it although I think I came closest to understand your thoughts in the similarity field) but is more on how you manage the risk and your money while trading these ideas. Thus, if you are not interested in how I perceived your theory, you can directly go to the end of this page and continue reading there (although I'd still need help with the other concepts).
May god bless you for your help and kindness!
D
Inhaltsverzeichnis
General idea
Eurusdd added supposedly later on a formal description of the similarity idea to the 1st post in the similarity thread:
This doesn't seem to be something special. It states that there are two stochastic processes that should be isomorphic (bijective?) most of the time. This imo could be either
- the price itself and an indicator that resembles the prices, or
- two indicators that show similar readings most of the time.
The Greek letter λ can then be interpreted as a parameter (i.e. indicator setting) that determines the relative frequency of similarity. Following Eurusdd's statements, λ should be chosen in a way that the two processes are similar most of the time (almost surely), i.e. > 90 % similarity (e.g. P = 0.97).
I interpret it this way: If two processes are hardly dissimilar then one could assume that a similar state follows a dissimilar state most of the time. Thus, the stochastic processes are somewhat predictable (at least if further properties are true, e.g. that a trend will most likely continue instead of reverse). Let's see how it does apply to the presented similarity setups discussed in the course of the similarity thread:
The CycleIdentifier application
The original idea presented in the similarity system thread was to use the cycle identifier on two different TFs and scale the settings (PriceActionFilter or Length) of the indicator among the TFs with the factor of the TFs, i.e. if I use Length = 5 in the M5, I should use Length = 5 × (TF5 / TF1) = 5 × 5 = 25 in the M1. Later on Madmoney adopted the scaling logic but applied two CI's in just 1 TF, i.e. he would have used in the M1 a Length = 5 and another CI with Length = 25.
In order to study the repainting behavior (which shouldn't be a problem, said Eurusdd), I decided to use Myxomop's non-repaint CI-based indicator that he posted in Madmoney's Ci/Symphonie extreme similarity eplained! thread.
The trading idea is simple: Most of the times the CI signals are identical after the close the the M5 candle. Thus, if there is a dissimilarity after the M5 close, it means that the price direction that causes the signal on the smaller TF indicator is highly false. Thus, the dissimilar CI spike gives a trade signal (if it is an up signal go short; go long if it is a down signal). Let's see how this looks like in the original 2 TF approach:
This is very far from being good. In the long run I would have won only 50 % of my trades.
So, let's see how it would have turned out following Madmoney's logic (trading on 1 TF and only take the 1st signal in the trend direction). The problem still is that a CI spikeis assumed to mark the end of a trend. However, in reality, the spikes are repainted quite often and thus I would have lost as often than I would have won.
I've seen in real-time how Eurusdd, Madmoney and Juhanimi made calls in real time. I've seen that they were highly successful and had losses only on rare instances. How was that possible? How could they have 100 and more trades without a single loss?
Is my understanding of the similarity idea a big misconception? I studied all setups presented by Eurusdd and Modmoney in detail. And they seemed to use the logic as described above.
... BUT HOW COULD THEY WORK IT OUT ...
and I can't although I study that shit for a half decade?
Applying the similarity idea to ZigZags
I never understood the logic or the messy code behind the CI/Symphonie indicator. However, I think that dissimilarities there are produced since they are maybe based on the close price (but I don't know for sure). Thus, a dissimilarity could appear if a close on the smaller TF is either the highest or lowest close within the calculation period, but the higher TF's candle closed lower/higher than the previous high/low.
In order to overcome that problem I switched to the ZigZag indicator. It is well known that this indicator repaints a lot. Thus, I've written again an indicator that shows me all formerly repainted as well as the final leg positions. Now if I apply the indicator with the scaling logic to two different TFs and scale properly, I won't see any dissimilarities:
Later on I realized that I even don't have to scale the settings according to TFs if I want to study dissimilarities in 1 TF. What would happen if I use two slightly different numbers for the Depth setting of the ZZ?
Answer: I will only see dissimilarities on rare occasions.
Since both ZZ indicator readings are similar in > 95 % of the time, I will only see in < 5 % of all dissimilarities that they are not taken out by a trend continuatio in the spike's direction. (One of those fails is the dissimilarity on the far left in the image above.)
My logic is this: If I use ZZ Depth settings that are relatively big, most ZZ legs will repaint most of the time. Since new trends will only start if two ZZ legs agree (almost surely), whenever I see a dissimilarity now, I can expect that price will continue in the dissimilarity's direction until a similarity appears. My ZZCallRepaintLegs indicator is modified and thus differs with the common MT4 ZZ in 1 point: A new leg is only drawn if a higher high or a lower low is established than the high/low of the last ZZ leg. Thus, my profit target is at least the high/low of the bar causing the dissimilar ZZ leg. Sometimes the profit target can be even bigger (but this depends on the history before the spike occurs).
Money/Risk management pitfalls
To me there seems to be one inherent truth that applies to all mechanistic trading strategies. This is that the potential loss always increases with the relative frequency of winning trades. That trade-off prevents me from being a successful trader.
From all my studies I think that this trade-off also applies tothe similarity trading approach (at least in the way I understood it).
The idea of similarity trading is to enter a trade once a dissimilarity occurs, with the expectation that the dissimilarity will be resolved with a similarity before the trend ends. However, that implies to exit whenever a similarity is observed. Hence, it doesn't matter if the similarity occurs if we are in profit or in loss. It just shows that a loophole in the market is closed.
From what I've seen in my painful studies, backtests and theoretical framworks is that in the long run I couldn't be successful if I follow that mechanistic workflow. At best, I will end up at break even. And even Eurusdd mentioned in his posts that one needs to be able to hedge for risk management.
- Thus, I'd like what the best strategy for hedging the similarity stuff is.
- Where/When/At which price develop should I start hedging?
- How/When/Where should I unhedge my position if the original trade was successful? Should I just rely in the transience of the markets? But if I opened my hedging position in the wrong place (the top or bottom of the market), I could wait for many years. In the meantime the costs could have eaten up all my profits ...
I pray that you could shed some lights on these questions and help me to become (after many years of pain) a successful trader.
Thanks a lot and may god bless you
D
