Саркисов_резюме_ENG (1138722), страница 2
Текст из файла (страница 2)
Classifications trees and neural networks were again estimated foreach stock taking into account new data. During rebalancing there was asale of securities of companies that were in the portfolio and were assignedthe class {sell asset} in the process of rebalancing. Funds received from thesale of securities were allocated among assets with the class {buy asset}.Weights of shares during rebalancing were determined by solving:max −= +∑=1 ̅̅̅ = √∑=1 2 2 + 2 ∑−1=1 ∑=+1 {(3),∑=1 ̅̅̅ = 1 − ℎ ≥ 0,where RHS assigned to shares with {hold asset} status. This share wascalculated as sum of producs of weights of certain shares and their return.During forming of a portfolio based on the Bayesian method ofkernel regression there was no need to assign a particular class todependent variables based on return. Rebalancing was conducted bysolving the problem (2), because Bayesian method doesn’t assume to selectassets to hold. In this case, each time the portfolio was formed based onthe expected return, estimated by the Bayesian method.The description of the variables used to form the input vector isbelow: Macroeconomic: GDP, GDP growth rate, inflation, inflationgrowth rate, bi-currnecy Basket value, trade balance, Brent’s oilprice Fundamental variables: current ratio, quick ratio, cash ratio,long-term debt-to-equity, total debt-to-equity, debt ratio, financialleverage, net profit margin, return on equity, return on assets,return on common equity , changes of EPS to share price, BookValue per Share / Share Price, FCFE, changes of FCFE, companycapitalization Technical variables: MA with different windows, Bollingerbands, momentum, Relative Strength Index (RSI), MovingAvereage Convergence/Divergence (MACD), Stochastic, tradevolume and Bid – ask-spreadUsing the optimal for each method input variable vector selectedfrom the initial set of variables investment portfolios were constructedbased on each of the three nonparametric methods.
The constructedportfolios showed higher profitability than market and portfoliosconstructed using parametric method throughout the entire investmenthorizon of 2016. The constructed portfolios were well diversified: onlywithin one rebalancing of kernel regression portfolio was formed, in whichthe weight of one share was 21%. Based on the results of this analysis thehighest profitability were shown by portfolios constructed using themethod of kernel regression.
In addition, nonparametric methods showedtheir high efficiency on 2008 data forecasting. At the same time, theconstructed portfolios showed a greater concentration than on 2016 data,but they were still diversified.Based on the results of the analysis of the most significant factors inthe construction of asset valuation based on nonparametric methods, it wasstated that in the construction of portfolios by all three methods, the mostsignificant variables for both periods (2016 and 2008) were: momentum, oilprice and value bid-ask spread. The obtained results confirm thehypothesis that the Russian stock market is speculative. Investors analyzethe movement of oil prices as a proxy for the general state of the Russianeconomy; try to invest in shares of companies that have already beenleaders in terms of past growth rates; ensure that these shares are liquidwith a minimum Bid-ask spread.In addition to solving the standard problem of maximizing income,which was formulated regardless to the preferences of the potentialinvestor, the task of maximizing utility for different values of theparameter of marginal risk aversion was solved.
Based on the results ofsolving the maximization problem, it was stated that use nonparametricmethods allows to obtain profitability not lower than the market rate forinvestors with risk aversion factors up to 16.01 for the method of kernelregression, 15.9 for the method of artificial neural networks , 15.4 - formethod of classification trees. The standard risk aversion factor varies from10-12 (Janecek, 2004), hence, use nonparametric methods allows to obtain ayield above the market for a large number of potential investors.
This factconfirms the effectiveness of nonparametric methods as a tool for forminginvestment portfolios on the stock market.At the last stage, the stability of the results was checked on randomwalk series. The following hypothesis was tested: main results of researchwere random and it is possible to obtain high profitability on random walkseries using nonparametric methods, which means that results didn’t basedon economic relationships. For testing this hypothesis random walk dataseries were created and algorithms created in this research were testem onthis data series.
50 different random walk time series of stock prices werecreated, “market index” were recalculated basing on this series and stockperfomances. According to the results of testing, no cases were revealedwhen use of non-parametric methods allows to obtain yield higher than themodeled "market". This fact indicates that the results of research are notaccidental and are based on economic interrelations between the variables.Main findings1. The necessity of using nonparametric methods as a tool forconstructing an investment portfolio was proved.
The efficiency ofthis type of methods was demonstrated by comparing theprofitability of portfolios formed using nonparametric methodswith the return of the market portfolio, and also the portfolioconstructed using the parametric method.2. It was shown that the main determinants of the Russian stockmarket are Momentum, Bid-Ask Spread, and the price of oil(Brent). This result is stable, because it was confirmed in theanalysis of all three nonparametric methods over both periods:2008 and 20163. It was proved that the portfolio formed using the method ofkernel regression had showed the highest return over timehorizons January 2016 - December 2016 and January 2008December 2008. The portfolio formed using the method of kernelregression also showed a strictly higher yield than portfoliosconstructed using other methods during solving the utilitymaximization problem for the investors with different riskaversion factors.4.
It was shown that during testing of the algorithms on random walkseries no positive results of the return was obtained , which provedstatistical significance of results.List of author’s original article1. Sarksiov A.R., Golodova J.G. Constructing of Investment Portfolioof Commercial Bank: Accounting of the Indicators of the Issuerson the Stock Market// Journal of Finance and Credit.
2012. № 35.pages 24-29. - 0.4 quire (personal author’s contribution – 0.3 quire)2. Sarksiov A.R., Bujanova E.A. Constructing of Optimal Portfolio onRussianStockMarketUsingNonparametricMethod–Classification and Regression Tree // Journal of CorporateFinance Research. 2016. № 1. pages 46-58. – 1 quire (personalauthor’s contribution – 0.6 quire)3. Sarksiov A.R., Bujanova E.A. Constructing of Optimal Portfolio onRussian Stock Market Using Nonparametric Method – ArtificialNeural Network // Journal of Corporate Finance Research. 2017.№ 3.
pages 100-110. – 1 quire (personal author’s contribution – 0.6quire).
![](https://s.studizba.com/z.php?f=/templates/si/i/noavatar.png&w=100&h=100&t=1"){kind=avatar}
