Virtual reality and a fjrms idiosyncratic risk: e-commerce case 1 - PowerPoint PPT Presentation

Virtual reality and a fjrm’s idiosyncratic risk: e-commerce case 1 Anna Loukianova , PhD (in Mathematics) Saint-Petersburg State University Ekaterina Smirnova Institute for Regional Economic Studies RAS 1 This research was conducted with the use of library and information resources of the Federal State Budgetary Educational Institution of Higher Education «Saint-Petersburg State University» A. Loukianova, E. Smirnova Virtual reality and a fjrm’s idiosyncratic risk: e-commerce case 1 / 40

1 Background 2 The Data 3 Model 4 Results 5 Discussion 6 Software applied & References A. Loukianova, E. Smirnova Virtual reality and a fjrm’s idiosyncratic risk: e-commerce case 2 / 40

Background Background Photo by Lucrezia Carnelos on Unsplash A. Loukianova, E. Smirnova Virtual reality and a fjrm’s idiosyncratic risk: e-commerce case 3 / 40

Background Virtual business… Virtual reality and a fjrm’s idiosyncratic risk: e-commerce case A. Loukianova, E. Smirnova retail-web#site/ analysis/on-target-rethinking-the- in (D. Gerdeman, 2018) a ) (S.M. Datar, C.N. Bowler (as cited . many companies are sufgering” companies, and it might be why of themselves as data to the point where they think have embraced data fully ”Even today, not all retailers Figure 1: width=2cm 4 / 40 a https://digital.hbs.edu/data-and- Photo by Zane Lee on Unsplash

Background textual features Virtual reality and a fjrm’s idiosyncratic risk: e-commerce case A. Loukianova, E. Smirnova analyses correlation and regression retailer’s operations on fashion e-commerce the impact of weather 2017) (Steinker et al., model a network min-cost fmow last-mile delivery 2016) (Wang et al., models with the use of E-commerce market challenges machine learning-based for the other consumers ’helpfulness’ assessment consumer reviews 2017) (Singh et al., literature review use in e-commerce big data analytics 2016) (Akter, Wamba, Model/ method Topic Study 5 / 40

Background E-commerce sales growth foresight “Volume growth in our US e-commerce channel in Q4 2019 was lower than our initial forecast.” “… we now have a fully operational e-commerce fulfjllment center for Rugs in the US and are expecting solid growth in 2020-21” — Thomson Reuters Eikon. (2020). [Balta Industries n.v. (2020, March 6).Balta FY 2019 Results [Press release]]. Retrieved March 6, 2020 from https://eikon.thomsonreuters.com/index.html A. Loukianova, E. Smirnova Virtual reality and a fjrm’s idiosyncratic risk: e-commerce case 6 / 40

Background What is the study’s aim? An algorithm development for a company’s sales changes nowcast adjustment for the company’s idiosyncratic risk Is approached through the objectives : defjne the idiosyncratic risk choose an approach to the idiosyncratic risk modelling train the approach on the real data A. Loukianova, E. Smirnova Virtual reality and a fjrm’s idiosyncratic risk: e-commerce case 7 / 40

Background What is the idiosyncratic risk The risk of the company’s cash fmows being afgected by the industry factors The idiosyncratic risk has the following attributes : the concept has emerged from the studies on the boarder of the market risk and the fjrm-specifjc risk management the idiosyncratic risk is traditionally assessed with the use of accounting data A. Loukianova, E. Smirnova Virtual reality and a fjrm’s idiosyncratic risk: e-commerce case 8 / 40

The Data The data source Data used in the study is from Amadeus Bureau van Dijk database. The companies from the industries “Computer programming, consultancy and related activities” and “Computer programming activities” were selected according to NACE Rev. 2 classifjcation (Source: Amadeus, Bureau van Dijk). A. Loukianova, E. Smirnova Virtual reality and a fjrm’s idiosyncratic risk: e-commerce case 9 / 40

The Data Companies’ country of headquarters Virtual reality and a fjrm’s idiosyncratic risk: e-commerce case A. Loukianova, E. Smirnova 10 / 40 Countries of Companies in the Sample Czech Republic Germany Netherlands Hungary Country Russian Federation United Kingdom Romania Sweden Belgium 0 5 10 15 20 Percentage

The Data Defjning the optimal number of clusters Virtual reality and a fjrm’s idiosyncratic risk: e-commerce case A. Loukianova, E. Smirnova 11 / 40 According to the silhouette method, the optimal number of clusters inside the sample equals 2 clusters. Optimal number of clusters Silhouette method 0.75 Average silhouette width 0.50 0.25 0.00 1 2 3 4 5 6 7 8 9 10 Number of clusters k

The Data market share: cluster 2 Virtual reality and a fjrm’s idiosyncratic risk: e-commerce case A. Loukianova, E. Smirnova Cluster analysis 12 / 40 ‘all the others’ have similar book value, but tend to have minor the ‘leaders’ have higher market share and tend to obtain higher book The two clusters in the sample evidence that software development market is low competitive: value: cluster 1 Cluster plot 10.0 7.5 Total Assets in 2017 cluster 5.0 1 2 2.5 0.0 0 2 4 6 8 Sales growth in 2017

The Data logistic Virtual reality and a fjrm’s idiosyncratic risk: e-commerce case A. Loukianova, E. Smirnova complicated than only a two-cluster model. This conclusion may indicate, that the real situation is rather more software development clusters are not homogeneous sales growth in bowth clusters follow primarily logistic distribution The distribution analysis summarising provided the following conslusion: logistic logistic BIC logistic AIC The sales growth values’ distribution gamma logistic Anderson-Darling logistic logistic Cramer-von Mises logistic logistic Kolmogorov-Smirnov Sales growth-2nd cluster Sales growth-1st cluster 13 / 40

The Data 0.07 Virtual reality and a fjrm’s idiosyncratic risk: e-commerce case A. Loukianova, E. Smirnova cluster. values. Each artifjcial dataset contained 10 thousand observations for each There were simulated 998 data sets of both clusters’ artifjcial sales growth 0.00 scale Imitation modelling 0.46 0.60 location Sales growth, cluster 2 Sales growth, cluster 1 with the parameters: The data was simulated using the logistic distribution for both clusters 14 / 40

Model Model Photo by Jared Murray on Unsplash A. Loukianova, E. Smirnova Virtual reality and a fjrm’s idiosyncratic risk: e-commerce case 15 / 40

Model (Genest, Ghoudi, & Rivest, 1995)) has the Virtual reality and a fjrm’s idiosyncratic risk: e-commerce case A. Loukianova, E. Smirnova all (𝑣 1 , …, 𝑣 𝑞 ) ∈ [0, 1] 𝑞 . copula or dependence function, defjned for continuous marginals where (𝑌 1 , …, 𝑌 𝑞 ) – a random vector with 𝑄𝑠(𝐺 1 (𝑌 1 ) ≤ 𝑣 1 , …, 𝐺 𝑞 (𝑌 𝑞 ) ≤ 𝑣 𝑞 )), (𝑣 1 , 𝑣 2 , …, 𝑣 𝑞 ) = following defjnition: model developed by A. Sklar (as cited in The econometric models applied Czado, Frigessi, & Bakken, 2009). Copula The simulation from pair copula model (Aas, The idiosyncratic risk-based model with the same parameters. as in the past (base) period growth would be distributed imaginary company’s sales implies that the average The “naive” nowcast, which The baseline model 16 / 40 𝐺 𝑗 (𝑦 𝑗 ) = 𝑄𝑠(𝑌 𝑗 ≤ 𝑦 𝑗 ); C – its associated

Model Idiosyncratic risk’s indicator Kendall’s tau (Kendall, 1938) was applied as the idiosyncratic risk’s indicator: 𝜐 = 2𝑇 𝑜(𝑜 − 1), where 𝑇 -the sum of the ranks; 𝑜 - the number of observations. A. Loukianova, E. Smirnova Virtual reality and a fjrm’s idiosyncratic risk: e-commerce case 17 / 40

Model where 𝑣 1 , 𝑣 2 − the two cluster companies’ sales growth; Virtual reality and a fjrm’s idiosyncratic risk: e-commerce case A. Loukianova, E. Smirnova 𝑙 - the quantity of the copula’s parameters. 𝜄 - the copula’s parameter (parameters); 𝑂 - the number of observations; 𝑗 - an observation index; ln [𝑑(𝑣 𝑗,1 , 𝑣 𝑗,2 |𝜄)] + 2𝑙, Copula model selection algorithm 𝑗=1 ∑ 𝑂 𝐵𝐽𝐷 ∶= −2 2019)) was used as the base for copula model selection: Akaike Information Criterion (H. Akaike, as cited in (Nagler et al. 18 / 40

Model Copula model selection Copula model selection results indicate Frank copula as the most preferrable by the algorithm on the base of AIC criterion. A. Loukianova, E. Smirnova Virtual reality and a fjrm’s idiosyncratic risk: e-commerce case 19 / 40 Simulation Modelling for a Model Selection Frank Model Rotated BB1 270 degrees Rotated BB6 90 degrees 10 25 40 Percentage

Model Sales growth simulation Virtual reality and a fjrm’s idiosyncratic risk: e-commerce case A. Loukianova, E. Smirnova 20 / 40 The sales growth simulated from Frank copula for two notional companies, representing two clusters. Frank copula was proposed by the AIC-based algorithm in most cases. 6 5 4 Pr 3 2 1 0 0.8 0.6 0.4 0.8 Company 2 0.6 0.2 0.4 0.2 Company 1

Model The Models’ Backtesting Virtual reality and a fjrm’s idiosyncratic risk: e-commerce case A. Loukianova, E. Smirnova 21 / 40 with the use of 2017-2018 sales growth data. Three copula models proposed by AIC-based algorithm were backtested 1.00 0.75 Sales changes adjusted 0.50 0.25 0.00 factadj naive BB1(270) BB6(90) Frank Weighted Model