The Impact of Stock Market on P2P Online lending Market’s activeness: An Empirical Study based on Chinese Peer-to-peer Lending Platform

1

The Impact of Stock Market on P2P Online lending

Market’s activeness: An Empirical Study based on Chinese

Peer-to-peer Lending Platform

Student number: s3497739 Name: Du Zhe

2

1. Introduction

In recent years, the internet finance has developed very rapidly in China, especially the P2P online lending market. The number of P2P platforms and the volume of transactions have been steadily increasing in China. Currently, there are about 1,700 P2P platforms in China, and the loan volume is nearly one trillion yuan in 2017 (RMB)1_.

P2P online lending has the characteristics of “high-yield” and “high-risk”, which is similar to the stock market that has long been favored by investors in China. Therefore, the p2p lending market is already becoming a new financial instrument market for Chinese investors. Many researchers study the relationship between traditional financial markets like the stock market and bond market, while less study for the P2P market. For this paper, we study the relationship between the stock market and P2P market.

Since the development of information technology, the development model of inclusive finance has no longer been limited to the traditional model of “Grameen Bank”, which originates from microfinance created by Muhammad Yunus. An innovation of inclusive finance has been occurred by means of a peer-to-peer (P2P) lending platform, which is the practice of lending money to individuals or small companies through online services that match lenders and borrowers. It has a flexible method and uses simple procedures that provide individuals with new financing channels and financing facilities, which is a useful supplement to the existing banking system. P2P online lending uses Internet technology to greatly improve lending efficiency and reduce borrowing costs. Compared with traditional bank lending, the P2P online lending threshold is lower and transactions are easier to operate. Moreover, on the one hand, China has a great number of small size entrepreneurs who need a loan but cannot fulfill the requirement of the bank loan conditions, and on the other hand, a growing Chinese economy produces a great deal of wealth for private investors. Therefore, the P2P industry has huge opportunities in China. In the Chinese investment market, housing market and the stock market were the primary options for Chinese household investors, but now the situation has been changed. Due to the housing bubble that happened these years, people prefer to put their money in the P2P lending market, which is convenient and with a relatively high yield. The average interest rate is about 12% and is higher than the traditional bank

3

loan rate (around 6%)2_{. So currently, the stock market and P2P lending market are the}

main investment market for Chinese investors. Is there any relationship between these two markets? By finding the historical record, in the year when the stock market changed drastically in 2015, the volume of transactions and the number of trading entities in these two markets rise one after another. The trading activity showed a phenomenon of “hot and cold alternating”, indicating that there may be some alternatives to these investment instruments. After the “Stock market crash “in June 2015, the overall transaction volume of the P2P online lending industry increased by 24% from the previous month3_{. To explain this phenomenon, in this paper, we study}

how the Chinese stock market impacts the activeness of the Chinese P2P online market. We find that many papers discussed the determinants of the probability of successful funding on the P2P online lending market, divided for 4 parts which are the loan characteristics, demographic attributes, social capital and loan characteristics. The main factors are borrower’s credit rating and financial strength discussed by Herzerstein (2008). Other authors suggest the demographic attributes like age, race, sex and appearance also have some impacts on the loan’s success. These factors are all internal environmental factors, but the study of the external environmental factors is rarely discussed, for example, the stock market. Many researchers studied the relationship between the stock market and bond market or gold market. Therefore, this is the first paper that studies the affect of the stock market development on p2p lending market investments.

For studying the correlation between stock market and P2P market, we will conduct an empirical analysis using multiple regressions. We collected the Shanghai index data from CSRC (China Securities Regulatory Commission), and the transactions’ record from creditease P2P online platform. Every transaction includes the loan characteristic, borrower’s attribute and identified information, and all the data are collected from 2014 to 2017. We find that the stock market has a negative impact on the P2P online market, especially the stock crash happens, its impact is more significant.

This study will consist of five main sections. The second section reviews the empirical study of P2P lending market and literature about the stock market and bond or gold market. The third section describes the details of data and its resource. Next, the fourth

4

section presents the result of regression and discussion. Finally, we conclude the paper by the fifth section.

2. Literature review & Back ground

In the past research literature, many authors like Charlotte et al (2005) and Smith (1992) they studied the impact of the stock market on other financial markets. They find that there are volatility spillover effects between stock market the bond market or gold market. Besides, many researchers like Herzerstein (2008) discussed the internal environment factors that influence the p2p lending market. However, there are little discussions about the impact on the P2P market from the external environment like other financial markets, especially from the stock market.

Since the activeness of P2P market depends on the amount of successful funding. Here we discuss the literature review about the determinants of the probability of successful funding. Many researchers studied the factor that influences the successful funding on p2p lending market, Herzenstein et al (2008) used the data from the Prosper online Lending Platform to study the influencing factors of borrowing success. They first divided the influencing factors into two categories. One is the characteristics of the loan, which mainly includes the borrowing interest rate, the borrowing amount and the bidding time. The other is the attributes of the borrower, including three types of categories, which are the demographic characteristics of the borrower (gender, age, race, marriage, whether there are children, etc.), financial status (credit level, income level, whether or not to own a house, etc.) and degree of effort (whether to join the group, describe the individual situation and the level of detail of the loan situation, etc.) The study suggests that there are multiple levels of relationships between borrower attributes, loan characteristics, and the success of the loan. First, the loan characteristics have a direct impact on the probability of loan; secondly, the borrower's attributes have a significant but relatively small impact on the success of the loan. In addition, the attributes of the borrower have a direct impact on the characteristics of the loan. The results show that compared with financial indicators and effort indicators, the demographic characteristics have little impact on financing outcomes. And compared with traditional financial institutions, the P2P network lending platform treats borrowers more equally, and reduced financial discrimination.

5

influences the probability of successful loan for 4 aspects, which are financial status, demographic characteristics, loan characteristics, and social capital factors. Firstly, regarding financial status, Peng et al (2015) pointed out this lending intention is mainly influenced by trustworthiness. Klafft (2008) studies the credit rating of the borrower plays a crucial role in the success of the loan, low credit rating borrowers can't get loans in the traditional banking system, and it is difficult to borrow on the P2P online lending platform as well. The author discussed the case of Prosper On the online loan platform, 54% of the loan application was initiated by the borrower whose credit rating is HR, but only 5.5% of the loan application was successful. However, those borrowing applications with credit rating AA-level borrowers, 54% of borrowing applications got funded. Iyer (2009) results show that the borrower's self-disclosed financial information (such as property information, personal income status, debt-to-income ratio, etc.) and the borrower's credit rating, default rate, overdue history and other information published on the platform will affect the lending behavior.

6

the loan, and persuade the lender to make investment decisions. Herzerstain et al (2011) find that the moral description is positively related to the future performance of borrowing, while the description of economic difficulties is negatively correlated with the future performance of borrowing.

At last, regarding social capital factors, Freedman et al (2008) find that there are loans with friends “endorsed” or friends bid, are fewer overdue rates, and have significantly higher returns. In addition, on most P2P lending platforms, members can form special groups spontaneously. If the group is formed because of the correct motivation, it can clear some information barriers.

7

haven than gold. These studies provide new ideas for us to study how the stock market impact on the P2P market’ activeness.

3. Methodology and hypothesis

On the basis of literature review, we expect that a decline in the stock market index will make investments in the P2P online market more attractive. Therefore, we make hypothesis in the following:

Null Hypothesis: The stock market has a negative correlation with the activeness of P2P market.

This study applies multiple regressions analysis to test how does stock market affect the P2P market in China. For finding the relationship between stock market and P2P online lending market, we need the historical records of stock market quotation and the borrower’s personal information and loan’s characteristics on the P2P online market. For the stock market, we can use the Shanghai index. Now we can build a multiple regression model:

𝐴𝑀𝑇

+𝛼

nnvv+𝛼

++𝛼

nnee+𝛼

vT+𝛼

gge +𝛼

+gT +𝛾

T

+𝛾

+𝜇 (Model 1)

vT, gge, and +gT respectively.

𝛾

𝛾

are

.

𝜇.

We give the explanation of variables as follows.

Explanatory variable:

Weekly index movement: nnvv

The shanghai index can explain well the Chinese stock market quotation, its movement can be measured as the change from last week to this week. So, we use Shanghai stock index as explanatory variable, which can represent the whole stock market average tendency level. And we choose weekly index movement, using the formula

ln 𝑖𝑛𝑑𝑒𝑥𝑡

8

where 𝑖𝑛𝑑𝑒𝑥𝑡 is the stock index of week at time t, and 𝑖𝑛𝑑𝑒𝑥𝑡−1 is the stock index at

time t-1. That also means the return of stock index every week. Since the investors observe the variety of the stock market and not only the absolute number on the billboard in the stock exchange.

Dependent variable:

eogarithm of gmount of weekly transactions: gvT:

For measuring the activeness of p2p market, we can see how many transactions happened in a fixed period. Here we use the total amount of transactions in every week from 2014 to the end of 20174_{. For keeping the same level with explanatory variable}

and interpreting better, we take the logarithm of the dependent variable.

Besides, the investors’ behavior is not only affected by the external environment, but also and mainly affected by internal environment. When borrowers put a loan project on the platform, they should offer many personal information and loan characteristics so that the lender can assess them and make investment decision. In the listing of characteristics, there are borrower’s purpose of the loan, the requested loan amount, the interest offered, the credit score, income level, the maturity of loan and other information like age education level. Moreover, the platform also encourages borrowers provide the house proprietary certificate and evidence of car ownership, profession, marital status, etc. Since this study is not studying the internal factors affected, our main aim is to test stock index factor, so we select the primary factors as control variables.

Control variables:

redit grade: +

The credit rating grade is a most representative indicator for lenders to assess how a borrower creditworthy is. It gives a large weight for lender’s decision. Berger and

4 _{Measuring the activeness can be done with many methods. The time for bidding the loan project is}

9

Gleisner (2009) studies the personal credit can impact the investor’s decision on the loan transactions. Klafft (2008) thinks Lower grade can get loan funded rarely not only on the traditional bank but also on the p2p online platform. It means that if there are many borrowers with higher credit grade, more transactions can get happened. Therefore, we expect the credit grade variable has a significant positive relationship with the amount of transactions.

nncome level: nnee

The financial strength affects directly on the borrower’s solvency, so borrowers’ income level can measure their average financial capacity. Herzestein (2008) and Iyer (2009) indicate that the financial strength plays an important role in the successful funding from P2P market. We use monthly salary as income level and we expect

nnee has a

eoan interest rate: +gT

The earnings of the investment of loan market are generated by interest and the quantity depends on the interest rate offered by borrowers. Puro (2010) finds that the impact of interest rates on the success of borrowing: the lower the base interest rate set by the borrower, the lower the interest rate of the loan, but the successful funding is also reduced. Based on this study, we expect higher interest rate attracts more investments and transactions during in one period.

eoan maturity: vT

The loan maturity is the duration of the loan, which means that when the borrowers can pay back the lender’s money. Normally, the lenders want to bring the money back as soon as possible that can invest in a new project and reduce the investment frequency. Chen et al (2009) indicate that the loan length has a significant impact on the successful funding of the borrower’s loan. We expect Loan maturity variable has a negative coefficient on the weekly amount of transactions.

gge: gge

10

to get loan. In our case, the main scope of age is under 50 years old (the more details of data will show in the data section), so without senior citizen, we can expect with more elder borrowers the amount of transactions would have a great number.

Dummy variable

ummy variable Trend: T

Building the multiple regression model only with control variables is not enough. As we know, the P2P market scale has increased rapidly, and a lot of P2P platforms have built in this decade. Thus, as the time effect would be also a factor that affects the activeness of P2P market, we also include a trend dummy variable, defined as DT.

ummy variable 2015-year stock great crash:

11

representative crash during the last decade, therefore, we include this crash as a special period event from June 12 to the end of August in 2015. Theoretically, since the capital is following profitability when the stock crash happens, the stock investors will withdraw the capital from stock disaster for risk averse and try to move the capital to another financial instrument that less risky and more stable. The P2P market could be an option to avoid the risk, thus we expect the crash dummy variable has a positive impact on the weekly amount of transactions on the P2P market.

First of all, I will construct a simple regression model just including the explanatory variable without other control variables to see the relationship between weekly amount of transactions and index movement. Then we build a multiple regression model by adding control variables step by step. We can see how the coefficient and significance of explanatory variable changes and how the variety of other control variables’ coefficient and sign. After that, we add the dummy variable both trend and stock crash dummy to see how the crash affects the P2P market. At last, we will test if this model has heteroscedasticity and autoregressive issues. By observing the line graph of the amount of transactions, we find it has volatility clustering in the period of stock crash. Indeed, the ARCH test shows it has ARCH effect. Hence, following the spirit of Hood M. et al (2013), we also build GARCH model in the section of the result.

4. Data

For the data of stock market, I collect the historical data of Shanghai index from CSRC (China Securities Regulatory Commission) annual report5 _{between January 2014 and}

December 2017, and I got the weekly index yield.

For the P2P market data, I collect the transaction data from the website creditease.com6_,

which is one of the largest P2P lending platforms in China founded in 2006. After years of development, the creditease loan business has covered more than 2,000 regions in more than 30 provinces across China, serving more than 2 million accurate users. The cumulative turnover of loans exceeded 46.2 billion yuan. Therefore, this platform is

5 _{http://www.csrc.gov.cn/pub/newsite/xxpl/sjspl/}

6 _{The founder used the summer to go to Bangladesh when he was studying America and followed the}

12

very representative and its data is convincible.

I obtain transaction data maturing between January 2014 and December 2017 by crawling the website of creditease. Through the website, 154872 borrowing requests were collected. In order to ensure the integrity and reliability of transaction data, after deleting the missing or inaccurate records, a total of 103860 valid transaction data were obtained. Based on these data, I got every loan project’s information and their categories such as successful borrowing requests, borrow amount, credit rating, maturity, income, age, loan interest rate etc. The transaction data collected in this article is all the full-scale loan information of the bidding completion status, which guarantees the integrity and reliability of the transaction data to a certain extent. Table 1 presents summary statistics for both dependent variable and independent variables. For every variable, we take weekly average level from 2014 first week to last week of 2017, and we obtain 209 observations in total. The statistics of variables such as mean, median and standard deviations, as well as maximum and minimum values show on the first row. The variables are reported in the first column from “weekly transactions” to “age”. In order to see the original data of weekly amount of transactions and weekly stock index, we also include them in the second raw and fourth raw separately in the first column. We can see the mean of the weekly transactions in the second raw, which is 473 and vary from 2310 at maximum and 11 at minimum. That means the range is very big, as can be seen from the standard deviation, reported in the last column, which is about 379.895. It might imply that the transactions are affected by some factors that cause this volatility, we can see this reason in the following section of the result. Because of its high volatility, we take the logarithm of amount of transactions, and we can see the standard deviation of variable gvT is decreased to 0.763, reporting at the end of third column. This is also a good way to reduce heteroscedasticity as we test it in the following section. Likewise, the weekly stock index also has a high standard deviation,

which is around 612.354 at the end of fourth raw. That is also the reason why we use the formula ln 𝑖𝑛𝑑𝑒𝑥𝑡

𝑖𝑛𝑑𝑒𝑥𝑡−1 to reduce the volatility and get weekly stock movement

13

Table 1

Variables Mean Median Max Min Std.Dv

Weekly transactions 535 473 2310 11 379.895 AMT 6.045 6.159 7.745 2.398 0.763 Weekly index 3045.703 3122.981 5166.350 2004.339 612.354 InMv 0.00217 0.003327 0.090735 -0.14290 0.0324 CR 174.793 176.479 180.038 150 6.261 InLe(yuan) 15201.800 12543.020 50000 2280.303 10580.560 RATE 11.089 10.857 13.245 9 1.422 MT(month) 30.604 30.964 38.223 25.143 3.064 Age(year) 38 39 42 28 3.755

Note: Table 1 presents the summary of statistics of dependent variable and independent variables. All their observations are 209. In addition, we also present weekly amount of transactions in the second row and weekly Shanghai index in the fourth row. For every variable, we report the mean, the median and standard deviations, as well as maximum and minimum values.

14

grade, and the credit grade range corresponding to each credit rating. The credit rating is divided for 7 levels from high to low are: AA (190-210), A (170-189), B (150-169), C (130-149), D (110-129), E (100-109), HR (0-99). Among them, the AA level borrower is rarely seen, which are just 24 borrowers have this level during these 4 years, as well as HR level borrowers, they just have around 1% percent of the total. The most of borrowers have A level, which is 96% of the sample during these 4 years. As we can see in the column of CR, the mean of weekly average of credit grade is 174.793, which is also belong to A level. Indeed, this makes sense, because all the transactions are full bid and successfully funded. Investors prefer to select borrowers who have a high level of credit grade, so every successful transaction may have high probability with high credit level. After the raw of variable +, it presents the raw of variable nnee, which reported the statistics of weekly average of income level. The income level is defined by borrower’s monthly income or salary and is also divided for 7 classes from high to low are 50000+RMB, 20000-50000RMB, 10000-19999 RMB, 5000-9999 RMB, 2000-4999 RMB, 1000-1999RMB, and under 1000RMB. The National Bureau of Statistics of China (NBS) 7_{has published the data of average income of Chinese workers: The}

average annual wage of employed persons in non-private units in cities and towns is 74318 yuan (monthly 6193 yuan), and the average annual wage of employed persons in urban private units is 45,761 yuan (monthly 3813 yuan). We can see the mean of the income level of borrower is about 15201.800 yuan, which is higher than the average wage of whole Chinese society. It implies that the borrowers on the P2P market have better financial strength compared with the national average income. The gap of income level is very big, as can be seen, the difference between maximum value (around 50000RMB) and minimum value (around 2280RMB) is about 47720RMB, which is also reflecting the economic inequality in China. The standard deviation of 10580.560 shows that there are quiet variations in borrower’s income. On the original data, it shows that the borrowers who have higher income level can make more amount of loan. This is just an inference from the sample data ostensibly, the further regression result is shown in the following section. The statistics of the variable of +gT is reported after the column of nnee. The interest rate offered by borrowers is not followed by borrower’s idea randomly, but set in a range by the platform. Chinese law stipulates that the loan’s

15

interest rate exceeds 4 times of the benchmark interest rate8_{(6%) of bank loans are usury,}

according to the current interest rate, it is about more than 24% annualized, which is not protected by the law. At present, the larger platforms will set their own interest rate caps within 24%, which is easier for most people to accept and easy to give people a relatively more standardized impression. For setting the range more specifically, the platform will further differentiate the interest rate range for each level based on the credit rating, the higher credit rating, the lower inferior limit. And then, based on the credit rating, the platform determines the final interest rate according to the different maturities. For example, if a borrower is rated as A, then the interest rates for 12 months and 36 months are definitely different. In short, the longer the maturity, the higher the interest rate inferior rate. But the relationship between the two is not linear certainly. At last, the borrower can set his own interest rate in the range followed by his own idea, for instance, if (s)he wants to attract lenders to invest him, he can set a higher interest rate. The mean of the weekly average interest rate is 11.089, which is very higher than the benchmark interest rate (6%) of bank loans. That is one of the reasons that P2P online platform can attract more and more investors to make investments on it. As we mentioned before, the interest rate is affected by many factors, it may have volatility during these 4 years, and it can be seen the standard deviation at the end of +gT column, which is 1.422. The penultimate raw shows the statistics of variable vT, which is the weekly average of maturity of loan. Normally, the maturity of loan is almost 2.5 years, which is about 30 months and corresponds the mean of the maturity, as can be seen in the second column of the penultimate raw, the longest and shortest maturity are about 3 years and 2 years respectively, which can be shown in the Maximum value and minimum value. It seems like this difference is not too big, but there is still some variation, as can be seen from the standard deviation, reported in the end of penultimate raw, which is about 3.064. The last variable is borrower’s gG , and its statistics show in the last raw. The mean of weekly average age is 38, which is similar to the median. Before, we mentioned that our scope of age is under 50 years old, and now we can observe that the oldest borrower is 42 years old, which means there is no senior citizen in our sample. And the youngest borrower is 28 years old, which means that even the youngest has the working capacity. To sum up, most of the borrowers in our sample are middle- aged, and few are in the student age, they are all in the working age. Therefore,

16

we expect that the age factor has a positive relationship with the amount of transactions, as the accumulation of individual wealth follows the age’s increase.

5. Result & discussion.

In order to guarantee the accuracy and feasibility of model estimation, we should need to test whether there is high multicollinearity between the main variables. The main problem with highly correlated regressors is that it may lead to high standard errors, rendering some variables insignificant.

To deal with these problems, first of all, we calculate the correlation between variables to see if these independent variables have multicollinearity. From Table2, we see that most variables are not highly correlated to each other, excepting +gT and vT, which a correlation coefficient of -0.5. Yet it is not so high. We focus on the second column, which is the correlation between variable nnvv and any other variables: they all have low correlations, under 0.5. Thus, multicollinearity is not a big issue for these variables.

Table 2 The correlation between the variables

Variables InMv CR InLe RATE MT AGE

InMv 1.000 -0.114 -0.325 0.097 0.044 0.026 CR -0.114 1.000 0.047 -0.396 0.224 -0.383 InLe -0.325 0.047 1.000 0.127 -0.366 0.262 RATE 0.097 -0.396 0.127 1.000 -0.517 0.085 MT 0.044 0.224 -0.366 -0.517 1.000 -0.251 AGE 0.026 -0.383 0.262 0.085 -0.251 1.000

Note: This table shows the correlation between every independent variable. All variables have 209 observations.

17

7(included all independent variables).

Table 3 OLS Regression Model

Step 1 Step2 Step3 Step4 Step5 Step6 Step7

InMv -4.517*** (-2.661) -4.348*** (-2.406) -1.408 (-1.359) -1.176 (-1.079) -1.158 (-1.095) -0.343 (-0.968) -1.162 (-1.042) RATE -0.039 (-0.640) -0.071 (-1.122) -0.027 (-0.380) -0.004 (-0.047) -1.181** (-2.512) 0.0089 (0.042) InLe -0.008*** (-1.358) 0.005*** (4.333) 0.002*** (-3.917) -0.294** (1.974) 0.001 (0.427) CR 0.024 (1.633) 0.023 (1.640) 0.001*** (2.737) 0.035*** (3.208) MT 0.0234 (0.888) 0.032 (-1.031) -0.041 (-1.579) AGE -0.027*** (3.457) 0.176*** (3.530) DT 0.010 (1.460) DC 0.959** (2.412) Adjusted R2 _{0.0321 0.0328 0.146 0.177 0.179 0.265 0.318} DW Test 0.897 0.896 0.997 0.930 0.924 0.912 0.974 White Test 0.803 0.364 0.162 0.000*** 0.000*** 0.000*** 0.015**

GARCH Regression model

Step1 Step2 Step3 Step4 Step5 Step6 Step7

18

Note: All the variables have 209 observations. The top half part is the OLS regression model with HAC errors, and the bottom half part is GARCH model. For brevity, the coefficients of monthly dummy are not shown. standard errors are in parentheses. *, ** and *** denote that an estimate is significantly different from zero at the10%, 5% or 1% level, respectively

First of all, we focus on the OLS regression model using the HAC9_{, as we can see in}

the table, the step1 shows a model just including the explanatory variable nnvv to see the simple relationship. As we can see in the following table 3, the variable nnvv hasa negative coefficient at 1% significant level, as well as after adding variable interest rate

RATE. It seems like our hypothesis is not rejected, but when we continue to add variable

income level InLe, the coefficient of our variable of interest is not significant anymore. But the coefficient of income level is very significant with a p-value at the significant level of 1% and continue to be significant until step 6. After adding variable credit grade

CR and other left control variables, arriving the step 6, we find that the coefficient of

variable index movement is still not significant. So, we can preliminarily think the index movement is not the main explanatory factor for the dependent variable, which means the variable InMv is not robust, but this inference should still have to verify after correcting the model. At last, we include every independent variable to build the model 1. After adjusting the standard errors, we get the result in the step7, which is shown in the last column in the table. We can see the variable InMv is not significant at all. Only the credit grade variable CR, the age variable AGE and the dummy variable stock great crash DC have a significant coefficient. The trend dummy does not have a significant coefficient, and this is not corresponding to the fact, which is the p2p market volume is developing rapidly. It seems like the other variables are not significant, and are inconsistent with the literature review that we mentioned.

In addition to the OLS regression results, we also present results using a GARCH model.

9 _{Table 3 shows the Durbin Waston statistic from step 1 to step 7. we find that all the Durbin Watson}

19

Because the amount of transactions has volatility clustering and the original model has an arch effect. The index of the stock market is financial time series, which usually do not have a constant variance of errors and may have volatility clustering, as mentioned by Chris (2014). Indeed, the 2015’s great stock crash is a turbulence case, which causes the highest volatility from 2014 to 2017. Now we can see if the time series of the transactions volume on the p2p market also has this kind of volatility clustering that affected by the stock market. Therefore, we conduct an ARCH. We choose 4 lag period, since the frequency is 4 weeks in one month as our variables are all weekly average level. Appendix 1 shows the result of the ARCH test. As we can see, the probability of the Chi-square test is 0.001, which is very significant. So, we reject the null hypothesis at 95% confidence interval, and we can conclude our model has ARCH effect. For resolving that effect, we should build a general autoregressive conditional heteroskedasticity (GARCH) model suggested by Tim Bollerslev (1986), which is more parsimonious, compared with ARCH model. We use the Maximum Likelihood method to define the lag of squared error term and the lag of conditional variance, and we get the GARCH (1,2) model that have least information criteria for the case of including all variables, but for other individual steps from step 1 to step 6, we get GARCH (1,1) model. Now we build the GARCH model (1,2) and for test its time effect to prove the robustness of dummy variable of the stock crash, we also include the monthly dummy from January to November, defined as D1 D2 D3 to D11 and the parameter to estimate is 𝑐₁ to 𝑐₁₁ :

𝐴𝑀𝑇_𝑡 +𝛿𝐷𝑇 +𝜃𝐷𝐶 +𝛼₁ nnvv+𝛼₂ ++𝛼₃ nnee+𝛼₄ vT+𝛼₅ gge +𝛼₆ +gT +𝛾₁ T +𝛾2 +𝜇+ 𝑐1 1+𝑐2 2+𝑐3 3+…+𝑐11 11 (Model2)

𝜎_𝑡2= 𝛼₀+ 𝛼₁𝑢_𝑡−12 + 𝛽₁𝜎_𝑡−12 +𝛽₂𝜎_𝑡−22 GARCH (1,2)

where 𝜎_𝑡2 _{is the conditional variance, 𝑢} 𝑡−1

2 _{is lagged squared error term at time t-1,}

and 𝜎_𝑡−12 _{is lagged variance term at time t-1, likewise, 𝜎}

𝑡−22 is lagged variance term

20

variable of interest becomes insignificant. Thus: the stock index movement doesn’t have a robust impact on the activeness of the p2p market. However, this situation has been changed in the last two steps. We arrive at the last step, we see each variable is significant at different significance level. The coefficient of variable nnvv shows that if other conditions remain unchanged, the weekly index movement is going up for 1%, the weekly amount of transactions will go down for 3.047%. Investors may transform their capital to the stock market when it comes bull market, and vice versa. Before we mentioned that the P2P market has developed rapidly in the recent years, now we can see the trend dummy has a positive coefficient hold a significance level of 1%. It corresponds to the fact that the volume of transactions on the P2P market has been steadily increasing in China. At last, we focus on the dummy variable of stock great crash DC, and its coefficient can also reflect the impact of stock market on the P2P market. It can be shown, the coefficient is 0.669 and hold a significance level at 5% level. It indicates that when the stock crash happened, some investors will withdraw the capital from the stock market and put more capital on the P2P market. We already discussed the variable of index movement is not robust, so can we wonder this situation could also on the dummy variable of the stock crash? For proving its robustness, we include monthly dummy from January to November, and we find that there is no significant coefficient between them neither of one month. It means that our model does not have a specific monthly effect, and the investors are more sensitive to stock crash compared with slight stock index movement. Because most of the P2P investors are not professional financial investors, they concentrate more on the borrower’s credit grade and income, etc, and may not focus on the stock index or analysis. So, some of them are highly possible to be aware of the extreme condition instead of slight volatility of stock index.

21

6. Conclusion.

This paper studies the relationship between the stock market index and P2P online lending market in China. We discussed the literature about the determining factors that influence the probability of successful funding and the relationship among the stock market and other financial markets. Then, we collected the transaction data with loan characteristics and borrower’s information from one of the largest Chinese Online P2P lending market creditease, which guarantees the applicability, authority for the empirical analysis. The data’s range is from January 2014 to December 2017, including the stock great crash in the year of 2015 on purpose, which is for testing how the stock crash impact on the P2P market. This is also the reason why we include the dummy variable of the stock crash. Then, we did the empirical research using both OLS model and GARCH model, including the index movement variable and other main internal factors as control variables.

22

23

References:

Berger, S. C., & Gleisner, F. 2009. Emergence of financial intermediaries

in electronic markets: the case of online P2P lending. Business Research Journal, 2(1), 39–65

Barasinska, N. and Schäfer, D. 2014. Is Crowdfunding Different? Evidence on the Relation between Gender and Funding Success from a German Peer to Peer Lending Platform. German economic review, 15（4），pp.436-452

Charlotte, C., Angelo, R., 2007. Realized bond-stock correlation: macroeconomic Announcement Effects. Journal of Futures Markets v27 n5 (May 2007): 439-469

Chen, Z., Li, B., Keung, G., Yin, H., Lin, C., & Wang, Y. 2009. How scalable could P2P live media streaming system be with the stringent time constraint? In IEEE International Conference on Communications (pp. 1-5)

Chris, B., 2014. Introductory Econometrics for Finance. Cambridge University Press, New York.

Chulia, H., Torro, H. 2008. The economic value of volatility transmission between the stock and bond markets[J]. Journal of Futures Markets, 28(11).

Freedman, S., Jin, G, Z., 2008. Do social networks solve information problems for peer-to-peer lending? evidence from prosper.com. NET Institute Working Paper No. 08-43; Indiana University, Bloomington: School of Public & Environmental Affairs Research Paper No. 2008-11-06.

Gonzalez, L., Loureiro, Y, K., 2014. When can a photo increase credit? the impact of lender and borrower profiles on online p2p loans, Journal of Behavioral and Experimental Finance, (2), pp. 44-58.

24

Determinants of success in online peer-to peer loan auctions. Bulletin of the University of Delaware, 15(3), 274–277

Herzenstein, M., Sonenshein S, Dholakia U M., 2011. Tell me a good story and i may lend you money: the role of narratives inpeer-to-peer lending decisions. Journal of Marketing Research,48(SPL), pp. S138-S149.

Hood, M., Farooq M. 2013 Is gold the best hedge and a safe haven under changing stock market volatility? Review of Financial Economics 22(2):47–52 · April 2013

Iyer, R., Khwaja, A. I., Luttmer, E. F. P., & Shue, K., 2009. Screening in new credit markets: can individual lenders infer borrower creditworthiness in peer-to-peer lending? Scholarly. Articles, 15242(rwp09-031).

Kanas, A., 2000. Volatility spillovers between stock returns and exchange rate changes: international evidence. Journal of Business Finance & Accounting 27(3&4):447-467 · June 2000

Klafft, M., 2008. Online peer to peer lending: a lenders’ perspective, proceedings of the international conference on e-learning, e-business, enterprise information systems, and e-government. 2008, H.R. Arabniaand A. Bahrami, eds, pp.371-375.

Matthew, H., Farooq M., 2013. Is gold the best hedge and a safe haven under changing stock market volatility? Review of Financial Economics, 2013, vol. 22, issue 2, 47-52

Peng, W., Hai, C. and Zheng, C., 2015. Exploring the critical factors influencing online lending intentions. [J] Financial Innovation， v1 n1 (201512): 1-11

Pope, D.G., Sydnor, J.R., 2011, What’s in a picture? evidence of discrimination from prosper. com, Journal of Human Resources, Vol. 46, PP53-92

25

Tai, C, S., 2007. Market integration and currency risk in asian emerging markets [J]. Research in International Business and Finance, 2007, (21).

Bollerslev, T., 1986. Generalized autoregressive conditional heteroskedasticity. Journal of Econometrics, 1986, vol. 31, issue 3, 307-327

Appendix:

Appendix 1: ARCH test with 4 lags

F-statistic 4.935435 Prob. F (4,200) 0.0008