Discovering similar crypto assets with SynergyCrowds platform
This article presents the idea of discovering groups of similar (or very different) crypto assets. Technically this is performed by clustering crypto markets with the SynergyCrowds platform. The clustering analysis is the result of a synergy between SynergyCrowds and Kaiko.
What is clustering ?
Clustering is a way of splitting a set of elements into groups/clusters of resembling individuals. So the aim is to group elements in such a way that objects in the same cluster are more similar to each other than those in different clusters.
To define clustering, we compare it with classification. The task of classification implies that we already know the categories for splitting our data (e.g. good vs bad, profit vs loss, etc.).
However, when clustering, all we know is we need to somehow split the data into several categories, but we don’t know them at the beginning of the process. We discover them only at the end.
The approach is to start with an initial set of entities and find a way for building groups that are as different as possible, but contain very similar entities inside each group.
Clustering example
Imagine a stadium full of people at a football game. For example, if we were to cluster all people from the audience into at least two clusters, here is what we would do. One obvious idea is to separate them based on the color of their clothes, as most participants would wear the colors of their favorite team. So taking each participant, judging the color and putting it into the appropriate cluster would finally yield into a split of the audience into two groups.
But for the same task, there are other possible approaches — like splitting them based on the gender or age of participants. So the idea is that in practice, the same set of entities can be split into groups by following different strategies.
Clustering the crypto markets
To apply the concept of clustering to crypto markets, first we need to look at the nature of crypto data. In our analysis this refers to price and time. Nowadays we have lots of data about the price evolution of different crypto assets, usually together with traded volumes for each time unit.
In a myriad of different digital assets, it always gets very difficult for one to evaluate all of them.
At SynergyCrowds we use the power of AI in order to clusterize the digital assets based on their price evolution.
The discovered clusters contain similar crypto assets in terms of price evolution over a time frame. Their similarity is computed not only by comparing prices at certain times, but also revealing possible dependencies in their behavior over that time frame.
The SynergyCrowds platform discovers three crypto clusters:
- Climb — cryptos with an ascending trajectory
- Immobile — cryptos with a flat trajectory
- Fallback — cryptos with a downwards trajectory
Every clustering analysis is performed against a reference asset, like: BTC, ETH, BNB, EUR, USD, RUB etc. The user can easily analyze the discovered clusters for all crypto assets traded against a reference asset by switching the analysis to a target reference asset of choice:
Value
The clustering task is used in portfolio optimization, which is one of the most important issues in asset management.
In decision making, the direct benefit for the crypto users would be for example to have the same strategy — for example when operating with crypto or for diversification, when optimizing their crypto portfolio.
The SynergyCrowds clustering app aims at helping all crypto users to better understand their crypto portfolios and improve their structure.
The clustering app helps users to cut through the noise of the crypto markets. Noise, in this case, is the price evolution of all markets (symbol pairs) and the target is to discover relevant clusters.
Knowledge
The objective of SynergyCrowds is to produce reliable knowledge that supports smart decision-making of crypto users. With this clustering app, the produced knowledge is the association of every crypto asset with one of the three clusters: Climb, Immobile or Fallback.
Knowledge is discovered from the data source to the produced clusters in a fully automatized manner, comprising no human intervention in the process.
Data
At time of writing, the SynergyCrowds clustering app uses daily VWAP crypto data provided by Kaiko:
- 3169 symbols;
- 51 exchanges;
- 3531 symbol pairs;
For each symbol pair, the app computes aggregated information such as the total volume traded for a certain pair and the volume weighted average price. For this, the app uses the volumes and prices for each exchange on which there is trade on that pair. The app checks a total number of 6090 exchange-symbol-pair combinations on a daily basis.
For each symbol the app also lists the corresponding project name. This is primarily to disambiguate the true identity of the symbol, since some symbols are used by more than one project, creating confusion.
Each clustering analysis for the day (‘d’) time granularity is performed on a 50 day time frame.
More details
See the documentation of the crypto clustering app at: https://synergycrowds.github.io/knowledge/crypto-clustering-analysis.html
Watch a short movie explaining how to use the Clustering & Crypto Portfolio Management apps: https://www.youtube.com/channel/UCrifCmD5GbPyhpc8nKsW_xQ
Development plan
SynergyCrowds is extending its cooperation agreement with Kaiko in order to provide crypto clustering analysis for smaller time granularity in the aim of building further relevance for crypto users.
Glossary of terms
Symbol: the exchange symbol of an asset. Example: BTC, ETH, EUR, USD etc.
Symbol pair: two symbols which are traded at least on one exchange, defining a market. Examples (using Kaiko notation): btc-eur, rlc-eth etc.
Time frame: the total period of time on which the app performs a distinct analysis.
Thank you for reading this article!
If you are passionate about clustering and feel that you can improve this app or have your own method, join the SynergyCrowds clustering Slack channel
Contributors: Darie Moldovan, PhD, Ionel-Mihai Motoc, Sergiu Cosmin