Applications Of The Probabilistic Parallel Planning In The Stock Index Simulation And The Family Financial Planning

The SIS（Stock Index Simulation）and the FFP（Family Financial Planning）are daily financial applications.However,they are done by professionals and they are difficult for non-professional investors.On the other hand,existing methods have many disadvantages.For example,the mathematical programming methods do not have good interpretability.Therefore,this paper constructed AI Planning domain models for SIS problem and FFP problem.It used a planning simulator to solve these problems automatically.This study provided comprehensible financial applications for non-professional investors.This paper proposed a general framework to solve some real-world financial problems.It was based on the PPP（Probabilistic Parallel Planning）.It could automatically generate plans to achieve multiple goals.We transformed the SIS and the FFP problem into AI Planning problems.We considered several factors in these problems.These factors included parameterized variables,state/action constraints,possible events and correlations.Meanwhile,our models were described by the RDDL（Relational Dynamic Influence Diagram Language）for PPP.They were solved with a PPP simulator called rddlsim.The experiments were divided into two sets.The SIS experiments used two stock datasets（the SSE 50 Index and the SSE 100 Index）.Since some time point,we simulated the movement of stock index in the next year.The solutions were compared with the real index movement using the IRBP_RDDL（Incomplete Stochastic Boolean Policy）as well as other three simulation methods.These simulation methods included the linear regression,the SVM classification and the LSTM neural network.We used the cross entropy,the least squares and the Pearson correlation coefficients as loss functions.In addition,the FFP experiments used the financial planning experimental report of the Money Tree software in July 2018and the comprehensive financial planning experimental report of September 2013.We used HP_RDDL（Heuristic Policy）to solve total wealth change,including the multi-stage and dynamic discrete-time structures.The solutions,on one hand,were compared with the existing experimental report data,including the minimum,maximum and average values;on the other hand,were compared with other solution policy results at runtime.The research results showed that our stock index simulation effects were closer to the real stock index movement.Furthermore,our SIS simulation was better than the regression and SVM method,and almost equivalent to that of the LSTM method.On the other hand,our FFP was compared with traditional methods in two aspects.The total wealth trends were basically the same,but the HP_RDDL was more efficient than other policies.Moreover,our method had an advantage of interpretability.For example,it did not need any manual intervention and parameters adjustment in the solution process.The interpretability was mainly based on the formal definition of the domain.It not only indicated how all kinds of variables affect each other in the SIS and the FFP,but also the plan solution presented a state trajectory of the simulation result.In general,this paper was more meaningful new attempt in the field of financial applications.