Analysis on the Influencing Factors of Commercial Financial Asset Allocation Structure Driven by Big Data

From the 1970s to the 1980s, due to lack of choice, the main asset management mode of most urban and rural households in China was simple savings deposits, and there was no problem in asset allocation. Later, the issuance of treasury bonds gave urban and rural residents another choice besides savings deposits. In the past, some families have shifted from a simple deposit financing mode to a combination of investment economy and deposit economy. At the same time, it also met people's demand for low-risk financial investment at that time, so there was often a shortage of treasury bonds issued, but compared with the investment demand of the whole society, the amount of government debt was far from enough. With the development of the country, the increase of population income and the improvement of population education level, the lack of government debt makes it difficult for Chinese households to meet their investment needs. With the rise of some financial institutions in the financial sector, if they want to raise funds, they must provide publicly recognized dividends, combined with the age and low-risk needs of investors holding funds at that time, as well as certain capital investment profitability, financial institutions attract emerging capital through insurance and wealth management. Therefore, in the 1990s, household insurance management became an asset allocation, providing more wealth options for Chinese households. At the beginning of this century, with the continuous improvement of residents'income, many families jumped out of the restriction of financial asset management investment and made physical investment. Nowadays, household wealth management is investing in more and more ways, including a combination of stocks, bonds, credit and real estate, to diversify risk and seek additional returns. As shown in Figure 1 below.

Figure 1.

Historical Development of Household Financial Asset Allocation in China

The development of artificial intelligence has increased the better allocation of Chinese households and increased the mobility and profits of Chinese households, but this does not mean that the development of artificial intelligence has not had a harmful impact on the allocation activities of Chinese households. China's development has also increased the risk of family resettlement in our country. First, China's household financial asset allocation structure has changed, the proportion of savings deposits in China's household balance sheet has declined, and the proportion of risky assets in China's household assets has increased. In addition, the financial laws and regulations are not perfect, and the industry is actually contradictory: Chinese families with the majority of Internet financial products will naturally face a greater risk of asset loss. In addition, in the early stage of development, artificial intelligence provides more convenient development channels for traditional banks, increases the ability of banks to receive funds, and improves the liquidity of banks. However, with the development of artificial intelligence, the banking industry stands out in the competition of many industries, and artificial intelligence financial products have more liquidity and profit advantages than the same type of banking products, thus increasing the cost of the banking industry, thus increasing the cost of banks. Liquidity risk banks, businesses and businesses are all affected by AI financing, so to develop themselves, the industries they rely on must compete with AI financing, reduce interest rates and increase product yields, all of which are true. Implementation of financing. Competition in the same industry has aggravated the systemic risk in the financial sector. With the participation of artificial intelligence, financial allocation can’t avoid systemic risk through asset allocation. The risk of domestic assets is increasing.

The development of the Internet has brought financial enlightenment and awareness of wealth to Chinese families. Due to the impact of the Internet on the economy, Chinese families have begun to change the structure of family businesses. However, passive access to information alone can’t meet the investment needs of families. Chinese families, especially those with greater weight in the distribution of family activities, should actively acquire financial knowledge, better understand their risks and asset allocation, and have a good understanding of asset allocation. Domestic necessity and risk tolerance. Only by understanding the characteristics of different assets can we balance the structure of family assets and meet the needs of families through asset allocation measures. Over the years, Chinese households have become more involved in the investment market, but more than half of Chinese households have only one investment product, and only about 10% have three or more investment products at the same time. Investment security will increase risk, and households with too concentrated assets can’t avoid risk. Chinese families need to flexibly adjust the asset structure, realize the diversification of family assets and reduce family risks. Financial institutions should also lower the threshold for product investment. In general, Chinese households are more familiar with large financial institutions than Internet institutions. However, the threshold of average investment and wealth creation of traditional financial instruments is too high, which limits the limit of ordinary families in China. Because of this problem, financial institutions are combining Internet technology with Internet technology to generate economies of scale, provide investment barriers and meet the needs of Chinese families. Internet financial institutions need to improve the accuracy of mobile Internet terminals, reduce the cost of Internet use, and increase the research and development of cutting-edge technologies. We will combine artificial intelligence, big data and other technologies with the consumption lifestyle and investment habits of different families, invest in online financial products in a targeted manner, and provide Chinese families with an adaptive decentralized investment approach.

In traditional regression analysis, the measurement error of independent variables leads to serious errors in parameter estimation of traditional regression models, which constitutes a false model. While traditional factor analysis allows the use of multivariate labels in the creation of latent variables and can handle measurement errors, it cannot analyze the relationships between factors. Using structural equation modeling, researchers can handle measurement error in the analysis while analyzing structural relationships among the latent variables. Therefore, the simultaneous equations in the structural equation model include the following two types of equation:

Measurement equation: (1)X=ΛXξ+δ\[X={{\Lambda }_{X}}\xi +\delta \] (2)Y=ΛYη+ε\[Y={{\Lambda }_{Y}}\eta +\varepsilon \]

A measurement equation is a system of equations representing the relationship between the variables X, Y, and η, ξ.

Structural equation: (3)η=Bη+Γξ+ζ\[\eta =B\eta +\Gamma \xi +\zeta \]

Structural equations are systems of equations that represent latent variables and relationships between latent variables.

Based on these two equation models and some model parameters, any of the structural equation model parameters can be calculated by an iterative solution process.

Proportion of structure. The relational structure of the block reports the relationships between the latent variables and their indices. Since the latent variable is not directly observable, its measurement is determined by other observables. The relationship between them can be expressed by this equation. (4)x1h=π1h0+π1hξ1+v1h\[{{x}_{1h}}={{\pi }_{1h0}}+{{\pi }_{1h}}{{\xi }_{1}}+{{v}_{1h}}\] (5)x2k=π2k0+π2kξ2+v2k\[{{x}_{2k}}={{\pi }_{2k0}}+{{\pi }_{2k}}{{\xi }_{2}}+{{v}_{2k}}\]

The block has the following preset data structure. (6)E(x1h∣ξ1)=π1h0+π1hξ1\[E({{x}_{1h}}\mid {{\xi }_{1}})={{\pi }_{1h0}}+{{\pi }_{1h}}{{\xi }_{1}}\] (7)E(x2k∣ξ2)=π2k0+π2kξ2\[E({{x}_{2k}}\mid {{\xi }_{2}})={{\pi }_{2k0}}+{{\pi }_{2k}}{{\xi }_{2}}\]

In structural equation modeling (as in other models), all data must be normalized to overcome differences in the corresponding coefficients due to different units for each variable. Especially in structural equation modeling, the hidden variables ξ₁ and ξ₂ are unknown, and canonical data are needed. Normalization means that a variable has a mean of 0 and a variance of 1. Mathematically expressed as follows: (8)E(ξ1)=0\[E({{\xi }_{1}})=0\] (9)E(ξ2)=0\[E({{\xi }_{2}})=0\] (10)Var(ξ1)=1\[Var({{\xi }_{1}})=1\] (11)Var(ξ2)=1\[Var({{\xi }_{2}})=1\]

At the same time, the specification is as follows: the residual v_1h and v_2k in each structural equation is independent of the corresponding latent variables ξ₁, ξ₂. Mathematically expressed as: (12)r(v1h,ξ1)=0\[r({{v}_{1h}},{{\xi }_{1}})=0\] (13)r(v2k,ξ2)=0\[r({{v}_{2k}},{{\xi }_{2}})=0\]

Therefore, the model is determined by the type of the structural equation: the residual and latent variables of the structural equation of each block are also independent of each other. The mathematical expression is: (14)r(v1h,ξ2)=0\[r({{v}_{1h}},{{\xi }_{2}})=0\] (15)r(v2k,ξ1)=0\[r({{v}_{2k}},{{\xi }_{1}})=0\] (16)r(v1h,ξ2k)=0\[r({{v}_{1h}},{{\xi }_{2k}})=0\]

In addition, we have to assume that the residuals of the structure equations of each block are also independent. The mathematical expression is: (17)r(v1h,v1h)=0\[r({{v}_{1h}},{{v}_{1h}})=0\] (18)r(v2k,ξ2k)=0\[r({{v}_{2k}},{{\xi }_{2k}})=0\]

Explanation: The mathematical definitions (14-16) are derived from the basic principles of the structural mathematical equation (A). The definitions (17-18) are not necessary for estimating partial least squares structural equation models. The above mathematical specification simplify that construction of the simulation data. At the same time, these definitions are also intermediate structural equation models that simplify two or more latent variables.

In structural equation modeling, the internal relationship represents the relationship between latent variables, which is a structural equation in mathematics. This internal relationship can be expressed mathematically as follows: (19)ξ2=β20+β21ξ1+ε2\[{{\xi }_{2}}={{\beta }_{20}}+{{\beta }_{21}}{{\xi }_{1}}+{{\varepsilon }_{2}}\]

The expected value of ξ₂ is: (20)E(ξ2∣ξ1)=β20+β21ξ1\[E({{\xi }_{2}}\mid {{\xi }_{1}})={{\beta }_{20}}+{{\beta }_{21}}{{\xi }_{1}}\]

As a corollary, there are: (21)r(ε2,ξ1)=0\[r({{\varepsilon }_{2}},{{\xi }_{1}})=0\]

Since in the inner equation of relation (19) there are ξ₁, ξ₂, ε₃ involved in linearity, the previous equations (12-13), (14-15) show that: (22)r(v1h,ε2)=0\[r({{v}_{1h}},{{\varepsilon }_{2}})=0\] (23)r(v2h,ε2)=0\[r({{v}_{2h}},{{\varepsilon }_{2}})=0\]

In order to study the influencing factors of AI in financial asset allocation structure driven by big data, 328 samples were collected, 316 valid samples were collected, and the effective recovery rate was 96.34%. The gender distribution of the sample is relatively even, with 188 males and 128 females. Most of the samples are between 25 and 40 years old, with a total of 245 people, which is more in line with the distribution of online financial management personnel. For investment, a certain level of knowledge is also required. 78.16% of the respondents have bachelor's degree, which is more in line with the official announcement of the investor's academic qualifications; From the perspective of occupation, 76.90% are white-collar workers, followed by civil servants, the proportion is 7.28%; 42.09% of the monthly income level is 5000-9999 yuan, followed by 10000-19999, the proportion is 40.51%. See Table 1 below.

According to the investment experience in Figure 3, the investment experience of the respondents is mainly collected between 3 and 4 years, accounting for 38.3%, the investment experience of one-year accounts for only 5.1%, and the Internet experience accounts for 97.8%, indicating that the respondents have financial experience, most of the respondents have rich investment experience and are relatively mature investors. These data show that the respondents are interested in money invested and manage finances, and have money invested and manage finances needs, which is consistent with the research background of this paper, so the research data of this study has a high reference research significance.

Figure 3.

Investment and financial management experience

Table 2 shows that the main investment purpose of the respondents is to manage personal or household funds. It is composed of investors' willingness to take risks, 12.3% of which are evasive, 75.9% of which are risk neutral, and 11.7% of which are preference. The frequency of investor activity is some times a month and some times a week, 42.4% and 32.3% respectively, which means that most of the respondents are short and medium terms investors. 98.1% of the respondents have heard of intelligent investment, but only 23.8% have used intelligent investment, indicating that the market acceptance and utilization of intelligent investment is still relatively low, which is in line with the financial habits of users. The results of the above studies are consistent. This shows that investors are not getting smart results from financial services, so it is necessary to study the potential mechanisms that prompt users to use them.

The mean value of shortage of accountability is 4.89 and the criterion deviation is 1.27, both of which are less than 5, that is, the sample's recognition of it is relatively low; the mean value of shortage of transparency is 4.42 and the criterion deviation is 0.91, that is, the respondents' recognition of the lack of transparency is relatively low; the mean value of personalized service is 5.57 and the standard deviation is 0.87, that is, the respondents' recognition of personalized service is relatively high; The mean value of perceived usefulness is 5. 27, the standard deviation is 0. 87, the mean value is greater than 5, and the criterion deviation is less than 5, that is, the respondents have a relatively high degree of recognition; similarly, the mean value of interest consistency, algorithm trust, service provider trust and willingness to use is greater than 5, and the criterion deviation is less than 5, that is, the respondents have a relatively high degree of recognition. Figure 4 below.

Figure 4.

Descriptive statistical analysis of variables

It can be seen from Figure 5 that the Cronbach's Alpha value of accountability loss is better than 0.7, which pass through the reliability examination; the Cronbach's Alpha worth of transparency loss is greater than 0.7, which also passes the reliability examination. Similarly, it can be seen that the latent variables are greater than 0.7, so there is good consistency, which passes the reliability test. The CR value of each potential variable is higher than the critical worth of 0.5, that is, the reliability of the model is higher.

Figure 5.

Reliability of the scale

If the structural equation model has many complex problems, it will lead to the distortion of the structural model estimation or the deterioration of the estimation accuracy. To check for multicollinearity, the VIF values are used for estimation. When the VIF value is less than 5, the collinearity problem is not considered to be serious. The results of the PLS operation are shown in Figure 6. The VIF of each latent variable is smaller than the critical value, indicating that the model does not have the problem of collinearity.

Figure 6.

Collinearity test

Figure 7 shows the results, all P values are less than 0.05 and all results are significant; the confidence algorithm's R ² value is around 0.67, indicating that its explanatory power is high; then, the service provider's trust R ² value is 0.604, and the willingness to use R ² value is 0.503. It shows 60.4% trust in the algorithm and 50.3% trust in the service provider, which is explained by lack of accountability, lack of transparency, alignment of interests, personalized service, and perceived usefulness. Cumulative variance variation explains 67.9% (R ² = 0.679), which has strong explanatory power, that is, 67.9% of users' intention to use is explained by trust in the algorithm and trust in the service provider. Overall, the model has strong explanatory power.

Figure 7.

Model R ² test results

Age, gender, innovation, privacy concerns and risk preference are selected as control variables. First of all, the control effect is analyzed to test which control variables have an obvious effect on the intent to use. The results are shown in Figure 8. Innovation and privacy concerns have an obvious effect impact on the intent to use, while age, gender and risk preference have no obvious impact on the intent to use.

Figure 8.

Test Results of Control Effect

According to Table 3 and Figure 9 of the running results, the lack of accountability perceived by users has an obvious negative affect on the trust of the algorithm (β = -0.167, P < 0.005), and the lack of accountability perceived by users also has an obvious negative affect on the trust of service providers (β = -0.125, P < 0.05). The lack of user-perceived transparency had an obvious negative affect on the trust of the algorithm (β = -0. 114, P < 0.05), and the lack of user-perceived transparency had no obvious direct impact on the trust of the service provider (T < 1.96, p > 0.05). Users' perception of alignment of interests has an obvious positive impact on trust in the algorithm (β = 0.153, P < 0.05), and also has an obvious positive impact on trust in the service provider (β = 0.115, P < 0.05). User's perception of personalized service not only has an obvious positive impact on algorithm trust (β = 0. 173, P < 0.05), but also has an obvious positive impact on service provider trust (β = 0. 269, P < 0.05). Users' perception of usefulness had an obvious active impact on algorithm trust (β = 0. 463, P < 0.05), and users' perception of usefulness had an obvious active impact on service provider trust (β = 0. 353, P < 0.05). There was an obvious active relation between users' trust in the algorithm and their intent to use it (β = 0. 518, P < 0.05), and between users' trust in the service provider and their intent to use it (β = 0. 206, P < 0.05).

Figure 9.

Direct effect test