Most of the content produced about crypto serves price prediction. Investors use the information they filter to determine which coin to invest in. Wouldn't it be nice if the information produced about cryptocurrencies was transferred to a forecasting model so that investments were made systematically? A large number of these models have already been produced, but we cannot say that they are widely used. VORTECS model offered by Cointelegraph to its paid subscribers is one of them.
Those who consider the price movements from the perspective of technical analysis come up with strategies that reflect a particular forecasting pattern. For example, buy if the RSI is above 60 and the MACD is also rising. It is also possible to test such strategies on historical data. Using simple rules derived from experience, a success rate of 60% to 65% can be achieved. (Provided that the correct stop-loss point and target price are determined) It is possible to access dozens of videos on Youtube where such strategies are explained.
If such a success rate is achieved by only making use of past prices, who knows how much our success rate will increase if we include other information. After all, it is possible to use RSI, MACD, moving averages, and many other variables together in a forecasting model. Before I go into more detail, I want to talk about the general framework for prediction models.
When we want to predict a variable, the first thing to do is to know the nature of the variable we are trying to predict. This work is called business understanding in the data analytics literature. It is not easy for an analyst unfamiliar with crypto to create accurate forecasting models.
Then, the data understanding phase is started. At this stage, the meaning, properties, and relationships of the data to be used for prediction are investigated. Then, the steps of data preparation, model building, evaluation, and implementation are carried out respectively.
By the way, I would like to point out that the model I am talking about is about predicting prices that will occur in a period of one day to one week. Longer-term prices are, paradoxically, easier to predict. Especially for Bitcoin and Ethereum.
The most important advantage of machine learning models is that they automatically determine which variable is decisive in the prediction. In fact, they also calculate the amount of influence of each variable. Therefore, the analyst who created the model feeds all the data she/he finds to the system, and the machine learning model finds the variables that best explain the target variable and calculates their weights.
It is possible to use machine learning models such as linear regression, deep artificial neural networks, and decision trees for the prediction of numerical data. I use linear regression in price prediction models that I am currently developing.
What variables other than technical analysis indicators can we use to predict the future value of a cryptocurrency?
Variables such as sentiment score, what percentage of inflation coin has, and the number of coins in the stock markets affect the results.
The models I have built so far have shown that the ratio of the trading volume in the last 24 hours divided by the total market value is also important.
So far, I've built models that predict prices for the next day and a week ahead. I included data from around 800 cryptocurrencies in the study. The last 1-hour price change is effective in estimating the price of the next day. It has a 20% positive correlation with the prices of the next day. In the weekly price forecast, how much the price changed in the previous week stood out as the most effective variable. Currently, I am testing the model I have built, in different time periods and I am doing various tests by adding new data to the model.
I know that the factors affecting the pricing behavior should be handled under two separate headings, permanent and cyclical. First of all, I try to find permanent (conjuncture independent) factors in the models I set up. I will deal with the cyclical effects in the second step.
After finalizing the model, I plan to trade crypto as per the model for a while. If I am sure that it works, I plan to present the model to the benefit of the community through the new coin I will produce on Hive Engine.
What would the ultimate price prediction model for cryptocurrencies look like? It should have had sufficient depth in terms of both cryptocurrency coverage and data diversity. It would determine the main trend that the relevant cryptocurrency is in, benefiting from support and resistance levels, momentum indicators, moving averages, correlations, divergences, and Fibonacci levels. Beyond the scope of technical analysis, it would consider information such as social media data, on-chain information, movements of whales. An ideal crypto price prediction model would take into account the macroeconomic situation and produce results obtained by compiling the information produced by several machine learning models. I aim to reach the ideal model gradually over time.
In 2021, I have benefited a lot financially from my intensive work on technical analysis issues. I have no doubt that my model trials will also have a positive impact on the financial results I will achieve. I hope it will go beyond that and turn into something bigger.
Thank you for reading.
Image Source: https://unsplash.com/