Cryptoeconomics: Pilot Study on Investments in ICO Startups Using Neural Networks

In the field of cryptoeconomics the Ethereum (Ethereum Foundation) project gave opportunity to create “own” cryptocurrency – new token based on its smart-contract platform to everyone without lowlevel programming skills. Then it became obvious that tokens could be used for crowdfunding as the Ethereum did in 2014. Unregulated and easy to access such scheme became popular among related to the blockchain tech startups. It was named Initial Coin Offering (ICO/or ITO). Despite its name, which is similar to IPO, this scheme is usually used for venture funding of a new project instead of expanding already well-established working business. The authors use machine-learning algorithms to classify ICOs and estimate ROI based on public digital data and web-sources. The goal of the research is to develop sustainable and efficient model, which will predict target profit ROI (profit trends) of ICO startup. Data collection and analysis period: Feb-Mar 2018. The prediction model and the application (service) of ICO startups’ selection are developed as the result of the study. Results. There were over 3000 samples of ICO-startups in the research dataset. After cleaning and elimination of outliers, it contained only 518. The number of samples with positive ROI (which means that these ICOs were profitable) was 234. Cross validation metric was confirmed to be accurate. The model achieved 79 % accuracy (average value). To prove this score separated prediction was executed the metrics: for test dataset AUC is 0.78; for profitable samples Precision: 0.76; Recall: 0.9 for profitable; F1-score: 0.82. Discussions. In order to achieve the objectives of this study, various IT components of the service architecture (applications) were developed to monitor, analyze and predict the risks of ICO startups. An artificial neural network was developed to solve the problem of ROI classification and prediction. The average ROI among profitable ICOS was 47 %. Taking into consideration that the crypto market is highly volatile and that there is a possibility that such investments will not bring any profit, this model of monitoring, analysis and prediction can be very valuable for the purposes of critical selection (exclusion) of a number of ICO projects from potential investment. Conclusion. The developed components can be used as a basis of monitoring service of ICO startups. The risk-forecasting model can be improved, foremost, by using the most complete (and wider) set of data. In this case, individual data collection and processing tasks can be performed manually, which will require additional resources. It should be noted that other types of neural networks can be developed for both text analysis and trading data analysis. This may lead to the logic of using a combination of models, which will potentially help to provide the most accurate predictions.