Beyond HODL: Can You Trade with On-Chain Data - TOKEN2049 Dubai 2024

yeah uh I want to talk about onchain data and specifically um the grand question whether you can use Onin data to effectively trade which I believe is a question for which there's still a lot of doubts nowadays and um I want to address this here today um spoiler alert you can successfully trade with Onin data so stick around and I show you some evidence for this now to start off with when we think about the value of onchain data I think that one use case that is widely recognized that probably all can agree with is um the usage of uh onchain for fundamental analysis right to get insights into core aspects of crypto assets um and their underline networks such as activity and usage adoption the distribution of a token across its holders um the health and the security of the underly network and so on and so forth right so it's a major use case actually many of our clients um use this use onchain data for um network uh or for fundamental analysis um but the question remains right so what about um the usefulness when it comes to Market related questions can we use onchain data for this um a glass note over the last last roughly six years we've gained a deeper and deeper understanding of um how to look at Onin data from a market lens and have adopted um our our framework and the the way on how we look at depending on on what we saw is most useful um with our clients and in our research and so um these are some of the major categories um that we these are some some of the major categories that we look at when we look at an onchain data from a market related perspective right so the analysis um of Supply Dynamics um the understanding of Market cycles and um using onchain data to help time tops and bottoms um understanding of investor Behavior such as long longterm holders short-term holders and other cohorts um onchain data um based pricing and valuation models um then profit and loss analysis which from an onchain perspective is incredibly use because you're able to assign an exact price aimed and Tim stamp to every single token and transaction on chain right so to make this a bit more concrete let me walk you through a few examples right um this one is an example of realized profit and loss right that we can use in order to detect um periods of extreme uh profitability taking um to identify potential local Maxima you can look at unrealized profit and loss in order to understand Market f basis potential over and undervaluations of an asset um we can analyze investor demand um by looking at um how many um old coins are moving into new hands at recent prices we can use a whole cohort of um Market cycle metrics in order to help understand are we close to a top or to a bottom um we can uh you know create various pricing models that help get a reference point for today's prize historically speaking and you know I could walk you through momentum Trend spending Behavior cell cide risk and whatnot um but um you might think those are nice charts nice visualizations um um but um you know it's it's a good set of cherry-picked um examples to make a point um not very scientifically arguably so I want to take completely different approach here and um do this from a very data driven perspective um to actually you know quantify how valuable onchain data actually is so here's what we do we create a machine learning model that we feed exclusively with onchain data dozens and hundreds of different onchain metrics um we trade and we we we train this model for a specific trading strategy in this case it's a very simple Bitcoin long only strategy meaning at the end of the day I'm either long Bitcoin or I'm flat right and the idea to do so is not necessarily just to obtain uh the model which in itself is something very valuable if you have a model that you can successfully trade on but what we aim in to do here is to go back to the training model um and dissect it in order to understand which are the data points which are the onchain metrics that most significantly contribute to the decision- making proc of this model right so we train the model we do cross validation hyperparameter optimization and the first thing that we need to assess is the model actually doing what it's supposed to do so here's its performance and yeah we see that the model works right uh you see the performance here in Blue uh and Bitcoin in Gray as a reference point um and it's a very conservative model uh right it protects from outside it catches the bull very nicely as well um the sta statistics show that um uh you know it is better than Bitcoin not in absolute uh terms of the performance right maybe not significantly but in some of the metrics that that matter most especially for institutional Traders such as risk adjusted returns as measured by the sharp ratio and the draw down which we're able here to reduce by 50% right and note that this performance is completely out of sample performance and life trading data right so there is no overfitting here this is exclusively data that the model has not seen before so um we have first evidence yes onchain is actually very valuable uh to improve trading decisions um but now we want to go back to to the model and actually inspect this we can do this because um the model is a tree based model so we can understand which are the features which are the metrics that most contribute uh to the decision-making process and this is the top list sorted by feature importance and we see things like the number of entities um in profit um mvrv which is a metric for unrealized profit and loss uh short-term holder soer which is a metric for realized so uh realized profit and loss um so we start to get this understanding that profitability metrics um are actually a useful set of metrics to look at when you're having this particular trading strategy um and not only which are the most importance but also what is the direction of the influence of a particular metric uh for the trading algorithm whether you know High values have a positive impact or a negative impact as you see colorcoded here um and now the question becomes can we actually drop the uh model uh entirely and use the information that we got here use the metrics directly in order to um you know improve our trading decisions so what we do is we go a level deeper in the dissection process we pick two of these metrics this is very arbitrary just for um presentation uh and illustration purposes and we'll look at their values and what they're saying for every single decision that the model is making um as you see here represented by each in every single dot right so you have you have these two metrics and now we starting to to get datadriven insides of which are the optimal ranges of values for each of these metrics that are most suitable to make um a call on a long position for for Bitcoin right so for short-term holder soer as you see on the sees it's very low values close to zero um indicating market equilibrium uh you know uh subsiding from investing or from going long in periods of high profitability or Panic selling and for entities in profit it's relatively High values potentially indicating positive momentum uh bull market but also look at the Nuance here right we're talking about relatively High values for entities in profit meaning that if you cross a certain threshold and you're too too much in profit right you get the risk of profit taking potential market downturns and so um we we we like the the interpretation of this data that the model is giving us is very very nuanced and very useful um so what we can do now right is we drop the model uh we use um the these two metrics just as an example we have this goldilock Zone that we identify as if you're here it's a good opportunity to go along Bitcoin and we create this very simple heuristic that we just using these two metrics where we say are we in the gold loog Zone meaning are these metrics um in their optimal range according to the model we go long Bitcoin and otherwise we stay out of the market um and we can run theistic and uh evaluate it again you see it here in red it performs extremely well right just using two onchain metrics again this is out of sample data and live paper Mone trading data right so um is there's no overfit or no no look Ahad byas um and looking through the statistics here right we're able to outperform Bitcoin in terms of absolute performance in terms of volatility uh we have a better risk return ratio and we're reducing the the the drw down significantly so um in conclusion uh what we see here is um that we're actually outperforming Bitcoin with just two unchain metrics right giving us a showcase that yes there is signal yes there is Alpha in Bitcoin that you can effectively use for trading right now um needless to say this is not Financial advice I'm not suggesting that you should be using this trading strategy uh to invest in Bitcoin right this is a very simple model um for the for the purpose of actually showcasing the power of onchain data um for trading um and um I think you know with a very with this very clear data driven approach uh we're able to do this um objectively we're running many more of these models um applied to different trading strategies in order to gain an understanding which are the metrics that help for other trading strategies independent of whether they're discretionary or systematic whether they're long long short and whatnot um and um are applying these across different crypto assets as well so if you found this interesting feel free to go uh visit this URL or talk to me uh reach out I'm more than happy to uh speak uh about this more um yeah thanks for your attention and uh enjoy the rest of the conference