Family life cycle optimal consumption decision-making method based on artificial intelligence algorithm

Abstract

The invention discloses a family life cycle optimal consumption decision-making method based on an artificial intelligence algorithm, and belongs to the crossing field of family finance and artificial intelligence technologies. The method can assist families in making long-span rational consumption investment decisions in complex and uncertain environments, and is another scientific and technical method for financial service families. The method comprises the following steps: 1, constructing a decision-making model under the diversity uncertainty of the family life cycle, wherein the decision-making model comprises the uncertainty of labor income, accidental expenditure and financial asset yield; and 2, designing an artificial intelligence decision algorithm fusing a UCT algorithm and Q-learning, and optimizing cross-period consumption and investment decisions. And 3, by improving node return value calculation and a Q value dynamic updating mechanism, rapid convergence of a high-dimensional complex model is realized, the problems of local convergence and poor expansibility of a traditional algorithm are solved, and the prediction precision of the optimal decision of the family life cycle is remarkably improved.

Description

Technical Field

[0001] The present invention relates to the cross - technical field of artificial intelligence and household finance. Specifically, it is a method for optimal consumption decision - making in the household life cycle based on artificial intelligence algorithms, which is applicable to household financial planning, insurance product design, and personalized wealth management. Background Art

[0002] In a complex economic environment, as an important micro - entity in the macro - economy, a household needs to make various decisions to maximize the total household utility within its life cycle, including consumption, investment, insurance, medical care, education, pension, etc. These decisions can generally be summarized as a decision combination of consumption, investment, and insurance.

[0003] In recent years, China's consumer market has gradually expanded, the financial market has been continuously improved, the types of consumption and investment tools have become increasingly rich, and the household consumption level has been gradually increasing. At the same time, more and more households have participated in financial markets such as stocks, funds, and insurance. In contrast, households are also facing increasingly complex financial uncertainties, such as fluctuations in future income, continuous investment in education and medical care, uncertain expenditures for unexpected events, the increasing burden of pension, and the volatility risks of the financial market itself.

[0004] How should a household choose between current consumption and future consumption, whether the allocation of various household financial assets is reasonable, and how to maximize the total household utility within the life cycle? This problem involves the inter - temporal decision - making within the household life cycle.

[0005] Traditional household financial decision - making models have two major defects: 1) Traditional household financial models are relatively simple. Nowadays, the financial environment is becoming increasingly complex and the number of uncertain variables is increasing. Due to insufficient consideration of variable factors, they can no longer meet the actual household financial decision - making needs; 2) Traditional household financial decision - making models use methods such as Newton's iteration method and recursive algorithms, which can only converge locally and cannot handle the scalability problem of complex uncertain decision combinations. Summary of the Invention

[0006] The purpose of the present invention is to solve the above - mentioned defects, construct a financial decision - making model that is more suitable for consumption and investment decisions in the household life cycle, consider the uncertainties of labor income, unexpected expenditures, and financial asset yields; adopt artificial intelligence technology, integrate the UCT algorithm and Q - learning algorithm, and develop a special - purpose solving algorithm for the model by improving the calculation of node return values and the Q - value dynamic update mechanism, realizing the rapid convergence of a high - dimensional complex household inter - temporal decision - making model, solving the problems of local convergence and poor scalability of traditional algorithms, and significantly improving the scientific nature of household life - cycle inter - temporal decision - making and the prediction accuracy of total household utility.

[0007] The present invention achieves the above object through the following technical solutions, including: Construction of a dynamic decision-making model: The decision-making goal in each stage of the family life cycle is to maximize the total utility of the family's entire life cycle. Using a family utility function in the form of CRRA, the decision-making goal can be expressed as Figure 2 A decision-making model under various uncertainties in the family life cycle.

[0008] Figure 2 In the formulae (1), (2), and (3), U represents the total utility of the family's entire life cycle; is the set of decision variables, , 1 ≤ 𝑡 ≤ N; and are respectively the proportion of consumption in disposable assets in the t-th period and the proportion of risky asset investment in the family investment; is the non-property income in the t-th period, reflects the consumption amount in the t-th period; is the family asset appreciation in the t-th period; , are respectively the risky investment return rate and the risk-free investment return rate of the family in the t-th period, is the family pension replacement rate, is a time-varying time discount factor, representing the degree of preference of the family for time in the t-th period. N represents the length of the decision-making cycle, and n represents the length of the income cycle.

[0009] Figure 2 In the formulae (2), is the family's unexpected expenditure ratio, is the family's disposable assets at the beginning, in formula (3), represents the insurance premium that needs to be paid after the family insures against unexpected expenditures, and k represents the minimum consumption ratio.

[0010] This model contains multiple decision variables and multiple stochastic variable parameters in each period, belonging to an intertemporal complex combinatorial decision-making problem.

[0011] Construction of an artificial intelligence decision-making algorithm: The decision-making algorithm is mainly divided into two parts: combining the UCT (Upper Confidence Bound Apply to Tree) algorithm with the MCTS (Monte Carlo tree search) algorithm to conduct simulation decision-making in the complex uncertain environment of model construction, and using the Q-learning algorithm for reinforcement learning and providing experience, and finally converging to obtain the optimal decision in each period and calculating the return value of each period's decision.

[0012] In the initial Select phase, the results of rapid simulation are used to make the current preliminary selection.

[0013] In the Expand phase and the Simulation phase, the accumulative family utility value (accumulative_value) from the initial state to the current state, and the accumulative family utility value (roullout_value) from the initial state to the final state are obtained respectively.

[0014] According to the trade-off ratio (m) of the family between the current income and the total future income, the expected return value (expected_value) of the current decision is obtained. The expected return value (score) of making a decision in the state is used as the environmental feedback information in the Q-learning algorithm, and the Q value is updated for reinforcement learning accordingly.

[0015] Figure 3 Artificial intelligence decision algorithm one, in formula (4), f(N) is an increasing function of the iteration number N, is the number of visits to the parent node, is the number of visits to the child node. c is the balance parameter of the UCT algorithm.

[0016] In subsequent loops, the results of the Q-learning algorithm's reinforcement learning are used as experience to guide the UCT algorithm to quickly simulate and make the current preliminary decision, and the expected return value calculated based on the current preliminary decision is used to update the Q-learning algorithm.

[0017] Figure 4 Artificial intelligence decision algorithm two, in formulas (5) and (6), is the state of the family in the t-th period, and m and n respectively represent the iteration numbers of the states 、 respectively.

[0018] The Q-learning algorithm will make the Q value tend to converge when it reaches a certain number of iterations. Description of the Drawings Figure 1 Structural diagram of the artificial intelligence decision algorithm Figure 2 Decision-making model under diverse uncertainties of the family life cycle Figure 3 Artificial intelligence decision algorithm one Figure 4 Artificial intelligence decision algorithm two The present invention has the following beneficial effects The present invention has the following beneficial effects

[0019] The present invention constructs a consumption and investment decision-making model for the entire life cycle of a family. This model incorporates more environmental uncertainty variables and decision variables, can more realistically simulate the actual situation faced by the family, and solves the decision-making model by developing a dedicated artificial intelligence algorithm to achieve practical guidance for the optimal consumption decision in the family life cycle.

Claims

1. The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings 1 and 2 of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present invention, several modifications and improvements can still be made, and all of these belong to the protection scope of the present invention.

2. Decision-making model under the diverse uncertainties of the family life cycle.

3. The present invention combines the experience of applying decision-making algorithms in household finance numerical simulation, and then constructs an artificial intelligence decision-making model for simulating the uncertainties of the family life cycle and supporting the optimal consumption decision-making of the family life cycle according to the characteristics and requirements of the household finance model and the application of the latest artificial intelligence decision-making algorithms. This model accurately and effectively constructs the constraint environment for families in a complex financial environment.

4. Decision-making model algorithm under the diverse uncertainties of the family life cycle.

5. The present invention integrates the artificial intelligence decision-making algorithms of UCT algorithm and Q-learning to optimize the intertemporal consumption and investment decisions, and realizes the rapid convergence of high-dimensional complex models by improving the calculation of node return values and the Q-value dynamic update mechanism, solving the problems of local convergence and poor scalability of traditional algorithms, and significantly improving the prediction accuracy of the optimal decision-making of the family life cycle.

6. The specific algorithm iteration process of the decision-making model under the diverse uncertainties of the family life cycle based on artificial intelligence algorithms is as follows: Before the iteration starts, obtain the family information and financial data when the family is created; obtain the Q-table data corresponding to the family. If the corresponding Q-table does not exist, initialize the Q-table, otherwise, directly use it.

7. (1) If the current state is the initial state of the family, generate the corresponding node root as the root node of the game tree; otherwise, directly use the node corresponding to the current state as the parent node; (2) Set the iteration number i of the UCT algorithm; (3) Use all legal choice sets (consumption and investment decision-making combinations) of the family in the next step (i.e., the second year) as the child nodes childs of the parent node in step (1); (4) Select a child node child∈childs to expand the game tree; (5) Quickly simulate from the selected child node child to the termination state of the family; (6) Traverse all the parent nodes of the current child node child backward, and update the expected return value and the number of visits of each parent node; (7) Update the Q value; (8) Go to step (4) until the iteration number reaches the set value i; (9) According to the number of visits of all child nodes childs, select the child node with the most visits as the optimal choice for the current state; (10) Go to step (2) until the current state is the termination state of the family; (11) Loop through steps (1) to (10) until the Q value converges.

8. It can be found from the above process that steps (1) to (9) are a process in which a family selects the optimal decision for the current state and simultaneously uses the Q-learning algorithm for reinforcement learning, where, The set iteration number of the UCT algorithm is limited.

9. Steps (1) to (10) are a process of selecting the optimal decision path in a household life cycle. It makes optimal choices using the UCT algorithm at each state of the life cycle. This process is continuous, and the optimal choices made under a finite number of iterations are not necessarily the globally convergent results.

10. Step (11) can be intermittent or continuous, so the number of iterations of the Q-learning algorithm is infinite. The rows of the Q-table are the corresponding periods of the household life cycle, and the columns are the consumption and investment decision combinations in the current state of the household. Then, after the Q-values converge, the trajectory formed by the columns where the maximum values of each row of the Q-table are located is the optimal path of consumption and investment decisions when the total utility of the household life cycle is maximized.

11. The present invention fully considers the characteristics and requirements of the household financial model, as well as the application of the latest artificial intelligence decision-making algorithms, and newly constructs an artificial intelligence decision-making algorithm for numerical simulation research on the uncertain decisions of the household life cycle.

12. The artificial intelligence decision-making algorithm used in the present invention can be compatible with the existing household financial models for numerical simulation of consumption and investment decisions in the household life cycle. Moreover, on this basis, it can also accommodate more financial variables. At the same time, it can continuously learn during the numerical simulation iteration process and converge to the optimal result faster.

13. Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Images