The impact of project characteristic on the ICO performance

0,43611

0,05678

0,5499

0,7772

0,00

1,33

Total

242

0,6689

0,43366

0,02788

0,6140

0,7238

0,00

2,09

It can be seen from the table that the average ICO success rates vary widely by project category. The most numerous category of projects is "Finance and Payments". This category includes payment systems, as well as projects on micro-crediting, investment, insurance, based on blockchain technology. The average share of ICO's success in this category was 0.7065, slightly higher than the aggregate average.

The second numerous category of projects is the blockchain services (47 out of 242 ICOs). This category includes projects that develop new protocols and platforms for blockchain, which do not relate to any particular area of activity. The average share of ICO success is below the aggregate average (Figure 2.11).

Fig. 2.11. Graph of average ICO success rates by project category

The most successful projects were the ICO from the categories "Logistics" (0.964), "Cryptocurrency" (0.8373) and "Trade and Advertising" (0.7928). Successes below the average for the aggregate are demonstrated by projects from the categories "Media and Social Networks" (0.49) and "Databases" (0.5825). The lowest ICO success is observed in the category "Games and stakes" - just 0.3938.

The variance analysis of the ICO success, depending on the category of the project, is provided in the Table. 2.12.

Table 2.12

Dispersion analysis of the ICO success depending on the category of the project

	Sum of squares	Degrees of freedom	Average square	F	Significance
Between groups	2,915	8	0,364	2,002	0,047
Within groups	42,409	233	0,182
Total	45,323	241

The table shows that Fisher's F- statistics is 2.002, and the significance is 0.047. Since this is below the accepted significance level б = 0.05, the null hypothesis about the equality of the means by groups is rejected. Thus, the average ICO success rates for different categories of projects really differ significantly from each other, i.e. the success of the ICO depends on the scope of the service being developed.

2.3 Building a multiple regression models for The ICO Success

The purpose of regression analysis is to construct a model that allows estimating the values of the dependent variable from the values of independent variables. The variables participating in the model are divided into two types:

1) the dependent variable Y plays the role of a function whose value is determined by the values of the explanatory variables that perform the role of the arguments;

2) independent (explanatory) variables X - factor attributes that affect the dependent variable.

The most common type of multiple regression models is multiple linear regression. Its parameters are determined by the method of ordinary least squares (OLS), the essence of which is to minimize the sum of the squares of the deviations of the actual values of the dependent variable from its calculated values.

To construct the multiple linear regression model, we use all explanatory variables whose impact on the success of the ICO was clarified by previous analysis - the price of the token, industry, the number of positive mentions on Twitter, the presence of a prototype, the bonus for early investment (Bonus).

Since a qualitative variable cannot be included in a linear regression model in its existing form, it was transformed into a dichotomous variable Industry1 with two values:

1) 0 "success below average" - this industry includes categories with ICO success rate below the aggregate average (databases, blockchain-service, Bets and gamble, media and social networks, etc.);

2) 1 "success above average" - the industry of projects with a share of ICO's success above the average (crypto-currencies, trade, and advertising, logistics, finance, and payments).

We use the "Forced inclusion" method and introduce all the variables listed in the model. The summary of the model is shown in Fig. 2.14.

Fig. 2.14. Summary for the multiple linear regression model

The criterion of the quality of the constructed model is the coefficient of determination R-square, which shows the proportion of variation of the resultant attribute, which is explained by the model constructed. In this case, the model explains only 14.1% of the total variance of the dependent variable, which is a quite low value.

The dispersion analysis of the model is shown in Fig. 2.15.

Fig. 2.15. Dispersion analysis of the multiple linear regression model

In the variance analysis table, the variance is calculated, explained by the regression equation (0.498) and the residual variance (0.169), as well as the Fisher's F-test. The value of the F-statistics is 2.955 with a significance of 0.016, i.e. The constructed model is statistically significant.

The coefficients of the model are given in Table. 2.13.

Table 2.13

Multiple Regression Model Coefficients

Model	Non-standardized coefficients	Standardized coefficients	t	Value
	B	Std. Deviation	Beta
1	(Constant)	0,524	0,095	-	5,540	0,000
	Token Price, USD	-0,013	0,012	-0,102	-1,044	0,299
	Presence of Prototype	0,206	0,087	0,238	2,370	0,020
	Positive Twitter mentions	0,000	0,002	-0,047	-0,464	0,644
	Bonus	0,326	0,141	0,232	2,302	0,024
	Industry	0,035	0,069	0,051	0,505	0,615

The multiple regression equation has the form:



This can be interpreted as following: with the increase in token price by $1, the ICO success rate decreases by 1.3% on average. If there is a prototype, ICO success rate is 20.6% higher on average. Each positive twitter mention increases the ICO success rate by 0 on average. If there was a bonus for early token purchase, ICO success rate is higher by 32.6%. If the industry that project is in is of those for which ICO success rate previously was above average - the predicted ICO success rate is 3.5% higher on average.

Standardized coefficients allow us to compare the effect of variables on the result. The greatest influence on the share of ICO success is the presence of a prototype (в = 0.238), then providing the bonuses for early investment (в = 0.232). The price of the token has a feedback with the result - the higher the price, the lower the share of the success of the ICO. The least influence on the dependent variable is the number of positive mentions on Twitter - its coefficient is almost zero, and в = -0.047.

The statistical significance of the regression coefficients is estimated using Student's t-test. In this case, the null hypothesis about the equality of the regression coefficient to zero is checked. If the significance of t is higher than the accepted significance level (0.05), then the null hypothesis is confirmed. It can be seen from the table that only 2 regression coefficients are statistically significant - with the variables "Presence of the prototype" and the bonus for early investment. However, the model, which includes only 2 dichotomous variables, will be uninformative.

Table 2.14.

The coefficients of the multiple regression model,

built by the method of step selection

Model	Non-standardized coefficient	Standardized coefficient	t	Val.
	B	Std. Error	Beta
1	(Constant)	0,517	0,059		8,764	0,000
	Industry	0,234	0,085	0,273	2,748	0,007
2	(Constant)	0,471	0,059		7,965	0,000
	Industry	0,247	0,082	0,288	3,002	0,003
	Bonus	0,386	0,134	0,275	2,869	0,005
3	(Constant)	0,392	0,066		5,941	0,000
	Industry	0,234	0,080	0,273	2,917	0,004
	Bonus	0,351	0,132	0,250	2,665	0,009
	Presence of prototype	0,196	0,081	0,227	2,420	0,017

It can be seen from the table that, as a result, the program has settled on an equation with three variables: industry, bonus and the presence of a prototype. The equation has the form:

This means that on average the probability of ICO success for projects classified as "higher than average" (logistics, trade, finance) is 0.234 higher than for projects from other industries.The provision of early investors with bonuses raises ICO success rate at 0.351 and the presence of a prototype - at 0.196.

The significance of the t-criteria for all three regression coefficients is below 0.05, which means that they are all statistically significant.

Fig. 2.16. Summary for multiple regression models,

built by the step-by-step method

Fig. 2.17. Dispersion analysis of the multiple linear regression model

The coefficient of determination for the constructed equation is 0.201, i.е. higher than in the previous model. It means that the regression model accounts for 20.1% of the variation of the dependent variable. The Durbin-Watson criterion is used to test the autocorrelation in the residues. Since it is close to 2 and above the threshold value (for the number of degrees of freedom k = 95 dU = 1.60), this means that there is no autocorrelation in the residues.

From Fig. 2.15 that the F-statistics for the equation was 7.691 with a value of 0.000, i.e. The constructed model is statistically significant.

For the purpose of further investigation, we constructed theLogit regression, which differs from the OLS regression model in that a dependent variable uses a dichotomous variable. In this case, we used the "ICO success" variable, which takes 2 values - 0 (ICO failure) and 1 (ICO Success) as the dependent variable. Logit regression estimates the probability of occurrence of an event using the formula:

where: z = b0+b1x1+b2x2+…+bmxm,

bi - regression coefficients;

xi - explanatory variables.

The results of the Logit regression are shown in Fig. 2.18. It was using step-by step method. On the figure we see that the model includes factors such as the price of the token, the type of the token, the number of Positive reviews, Industry, and Bonus for the early purchase.

The equation has the form:

Fig. 2.18. Logit regression coefficients

All coefficients are significant on 0.05% significance level.

The Model's coefficients cannotbe interpreted individually but overall model works as following:

let's carry out a forecast of the success of ICO on the basis of the obtained model. ICO of Debitum Network is a project in the field of finance and payments (Industry = 1), the price of the token was $ 0.26, the number of positive reviews was 36, the type of the token was Utility (Type = 1), the early purchase bonuses were 0.05 ETN (Bonus = 1). Then:

Z = -1.334 * 0.26 + 3.914 * 1 + 0.028 * 36-2.314 * 1 + 2.153 * 1-

-1.141 = 3.285

The probability of success of the ICO is:

those. the ICO most likely will be successful. Indeed, the success of the ICO was 1.

We can measure the quality of the Logit model is estimated using NagelkerkeR2 (pseudo-R2) , which is an adjustedversion of the determination coefficient for the binary model. The quality indicators of the constructed model are shown in Fig. 2.19.

Fig. 2.19. Quality indicators of the logit model

Nagelkerke R2 for the model was 0.496. This value suggests that theconstructed model explains about 50% of the variance of the resultant trait. This is a better indicator than the determination coefficient for the ordinary least square(linear) regression model, i.e. The Logit model better describes the original data.

Also, the quality of the model can be estimated using a classification tablewhere if obtained probability for the project from the model is closer to 0, we say that there is ICO Failed, if closer to 1 then ICO success and we compare it with actual results for those projects, in which the percentage of correct predictions is calculated (Table 2.15).

Table 2.15

Classification table for Logit regression

Observed	Predicted
	ICOSuccess	Percentage of correct
	ICO Failed	ICOSuccessful
Step 1	ICO success	ICO Failed	28	8	77,8
		ICO Successful	8	28	77,8
	Total Percentage			77,8

It can be seen from the table that the total percentage of correct predictions for the Logit model was 77.8%, which indicates a good quality of the constructed model and the acceptabilityof using it for forecasting purposes.



CONCLUSION

ICO or Initial Coin Offering is an act performed by the company or a group of people in order to attract financinginto a project which is based on the use of blockchain technology. Attracting of funds into the project occurs through the sale of tokens issued by the project itself in exchange for cryptocurrency or regular currencies (non-crypto). In fact, most of the tokens carry specific purpose and can be exchanged for some goods or services provided or created by the issuing company. After the ICO project's tokens are usually circulated on the exchanges (can be sold and bought by third parties).

In this work, a statistical analysis of the factors of the success of the funds was carried out by issuing and selling tokens. We used such types of analysis as correlation, regression, dispersion, and analysis of averages.

For the analysis, a sample of 243 projects that the ICO consistently conducted in 2017 was used. It was suggested that the success of the ICO is influenced by factors such as the price of the token, the type of the token, the number of accepted currencies, the number of the project team, the opportunity for the investors to pay with the currencies, and so on.

As a result of the correlation analysis, we found out that there is a weak negative correlation between the token price and ICO success. Thus, the issue of a cheap tokens increases the likelihood of successful attraction of funds.

Also, a correlation analysis was made of the relationship between the success of the ICO and the references to the project on Social Media. In order to do so, we created a script using Python Programming Language that allowed to make a sympathetic analysis of posts about the projects. The analysis showed that between positive references in social networks and the success of ICO there is a direct correlation of weak strength. The relationship between the success of ICO and negative or neutral statements on Twitter is not revealed. There is also no correlation between the success of the ICO and the number of team members, the number of accepted currencies, and the ability to pay in fiat currencies.

The relationship between the success of the ICO and the dichotomous variables was verified by means of averages analysis. As a result, it was clarified that the success of the ICO is influenced by the presence of the prototype and the bonuses for the early purchase of tokens.

The impact of the project category on the success of fundraising through the sale of tokens was assessed using variance analysis. The analysis showed that there is a difference between the success of ICO in different spheres of the economy. ICO in the sphere of trade, advertising, logistics, and finance is on average more successful than ICO in the field of media, social networks, databases, and sports betting.

To predict the success of the ICO, a multiple regression model was constructed. The model is based on three variables - the project has a prototype, bonuses for early investments and the industry in which the project will operate. However, since these variables only indirectly affect the success of the ICO, the accuracy of the constructed model is not high - the coefficient of determination was only 0.201.

That means that while conducting further investigation for the main causes of ICO success, the model can be improved by adding more significant regressors (now its limited to the collected data) and/or by changes of the model specification that in turn may improve the goodness of fit and explanatory power of the model.

Then, we constructed the Logit model that was found to explain about 50% of the variance of the resultant trait which is a much better indicator and model proved that it indeed can be used for the purpose of forecasting the ICO outcomes.

Those modelsmightbe used by crypto-investors and enthusiasts to predict the chance of the ICO success. Moreover, it can be used by agencies in order to rate the company's chances to successfully make trough ICO stage and if further developed accounting for the future project's performance it may be used to make the long-run investment decisions on those projects. But still, it should be taken into account that in past years there is rapidly increasing number of projects that “die” on post-ICO stage (they fail to operate successfully after getting required funds or they are just fraudulent). So, the model can be developed in future to account for that and predict not only the success of getting funding through ICO but to predict whether the project will succeed in later stages or not.

Finally, it can be used by companies that are planning to attract funds through ICO as they can increase the chance of successful fundraising by focusing on the factors that were found to be significant in the model and making the relevant improvements.



LITERATURE

Akst R. 7 sekretov. ICO. Ili pot u storonu tokenseila/R. Alst. - «Izdatel'skie reshenia», 2018..

Brigham U., Ehrhardt M. Financial Management Theory and Practice 10th Edition

Dougherty C. Introduction to Econometrics. Oxford University Press, 2011 (4th edition).

Dougherty C. Elements of econometrics. Study Guide. University of London. 2016.

EY research: initial coin offerings (ICOs), December 2017

Kondratova S. V., Umrihina M.V. IPO kak istochnik finansirovaniya deyatel'nosti kompanii. // Nacional'naya associacia uchenykh (NAU), 2015. - №7, P. 109-113

Laurent L. Blockchain: La rйvolution de la confiance /Translated from fr to Russian by A.N. Stepanova - M.: Eksmo, 2018.

Novikov A. I. Ekonometrika: Uchebnoe Posobie. - M.: Dashkov I K, 2013.

Swan M. Blockchain: Blueprint for a New Economy: Melanie Swan / M. Swan. - O'REILLY, 2015 (First edition)

Tikhhomirov N.P. Metody ekonometriki i mnogomernogo statisticheskogo analiza: Uchebnik. - M.: Ekonomika, 2011.

Yakovlev V.B., Yakovlev I.V. Klassifikaciya I snizheniye razmernosti dannyh v SPSS: Uchebnoye posobie. - M.: Editus, 2016.

Yakovlev V.P. Ekonometrika: Uchebnik dlya bakalavrov. - M.: Dashkov i K, 2016.

Размещено на Allbest.ru

...