Machine Learning Methods Implementation in Bitcoin Market Forecasting

The objective of my research is development and implementation of a forecasting model for the dynamics of a financial asset`s prices in the cryptocurrency market. Primarily, I need to evaluate the efficiency of various trading strategies, taking profitability for a quantitative measure of model quality assessment. Then, to develop a program that will trade using the most efficient of the selected algorithms. Finally, I need to calibrate my trading bot and test it in various market conditions. Applied methods are analysis and study of articles on trading in financial asset markets, monitoring the dynamics of asset prices and charts of selected indicators, as well as the analysis of expected signals to enter or exit the trading floor, evaluation the quality of the models are taken as a basis when building strategies. As a result, I expect to receive information about the efficiency of various trading strategies obtained empirically, then make a predictive model using machine learning methods and implement it to my program. For this purpose, in the course of work, I will develop a bot that will download price statistics from the trading platform, make decisions about buying and selling assets and writing them to a local database, then displaying the results of the actions taken. Further, the results will be analyzed, and the dataset will be used as a teacher for the predictive model using machine learning. My work contains the introduction, including problem statement and background, the main part with literature review, applied methods and results achieved. Ending up with a conclusion, describing tasks are still to be done, and appendices, where the program code is submitted. Relevance of my research is dictated by the attractiveness of the cryptocurrency market for investment due to low barriers of entrance and the absence of taxation. Trading in this market is very dependent on the accurate analysis and exact forecasting of significant changes in the value of assets using modern methods of data analysis. The popularity of using machine learning in the financial sector is due to many factors, namely, the ability of algorithms to classify objects, build regressions, clusters. Moreover, algorithms can perform similar tasks both having a pre-training data set, i.e. answers to tasks, and dynamically adjusted to changing data without prior training. Using machine learning algorithms can reduce costs and increase forecast accuracy with correctly selected data processing methods. Problem Statement. The object of study is the temporary financial series. The subject of the study is defined as predicting the direction of changes in time series using machine learning methods. The purpose of my research is development and implementation of a forecasting model for the dynamics of a financial asset`s prices in the cryptocurrency market. Primarily, I need to evaluate the efficiency of various trading strategies, taking profitability for a quantitative measure of model quality assessment. Then, to develop a program that will trade using the most efficient of the selected algorithms. Finally, I need to calibrate my trading bot and test it in various market conditions. For this purpose, in the course of work, I will develop a bot that will download price statistics from the trading platform, make decisions about buying and selling assets and writing them to a local database, then displaying the results of the actions taken. Research Objectives: 1. To get acquainted with sources on this subject 2. Select pairs of analyzed exchange indicators to identify entry and exit points from the market 3. Write a program that can perform the following actions: a) To upload prices of an asset from the trading platform for a given time period b) To calculate the values of the selected indicators, making decisions on the purchase or sale of assets based on their intersections c) To write all the decisions were taken to the local database and to display the profitability of the given strategy 4. To evaluate the profitability of each algorithm in different market conditions. To draw conclusions about the effectiveness of the selected indicators and the specifics of the application of the algorithms written on their basis Practical relevance of using machine learning for financial forecasting can be described by the quote of Babak Hodjat, a computer scientist who helped to lay the groundwork for Apple's Siri, “It is better to rely on what the data and statistics are telling you, rather than human intuitions and justifications” (Satariano, 2017). According to him, humans are too emotional for the stock market, they make too many mistakes. That`s why about 30 percent of institutional trading on exchanges takes place using algorithms.