Premium
Business cycle prediction: Application of Markov chain to online crowdlending
Author(s) -
Krishnan Sudeep,
Ashta Arvind,
Babu David
Publication year - 2021
Publication title -
strategic change
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.527
H-Index - 16
eISSN - 1099-1697
pISSN - 1086-1718
DOI - 10.1002/jsc.2428
Subject(s) - exponential smoothing , markov chain , autoregressive integrated moving average , computer science , econometrics , benchmark (surveying) , mean absolute percentage error , asset (computer security) , loan , statistics , time series , artificial intelligence , mathematics , machine learning , economics , artificial neural network , finance , computer security , geodesy , geography
Using Markov chains can improve the forecasting of the state of the total loan amount for the next month (growth, stagnation, or decline), compared to traditional forecasting techniques, if each of the previous month's basic information is available. Traditional statistical techniques have high forecasting errors and low accuracy for predicting the loan amounts and are no better than the naïve method of “no change” if data cannot be rapidly actualized. With recent data available, some traditional statistical techniques work better than the naïve method, but Holt double exponential smoothing has a higher mean absolute percentage error (MAPE). For predicting “states,” some traditional statistical methods improve with the degree of actualization (naïve, simple exponential smoothing), Holt and ARIMA decrease in performance, and TBATS remains the same. Since Markovian chains are better than all traditional time series forecasting techniques, it raises the benchmark for evaluating value added by artificial intelligence techniques for forecasting.