# Application of BSDE in Standard Inventory Financing Loan.

1. IntroductionBecause the small and medium enterprises (SMEs) do not have enough credit rating, there is no real estate or a third party to guarantee security. These factors lead to the financing difficulties of SMEs. To solve this problem, we can find an effective solution that is chattel mortgage loans to SMEs. Inventory financing business concerned in this paper is a class of chattel mortgage lending business in the logistics and financial innovation. In this business, the enterprise will provide its production, inventory, and other movable properties to the logistics and warehousing enterprises with legal qualifications in order to receive short-term loans from banks. However, the study about inventory financing business stays in the qualitative stage in many ways. Particularly the LTV ratios (ratio between loan amount and collateral value) are very important in this business. But banks depend largely on the experience of valuation in practice. These valuation methods of banks can not make quantitative analysis about the following factors, such as price volatility of the collateral, probability of default, loan maturity time, and marking to market frequency. Thus banks can not determine accurately LTV ratios of inventory commodities according to the banks' different risk tolerance levels. Therefore, our research about quantitative models of the LTV ratios will provide a scientific basis for the bank decisions and will have significant value of practice.

To get a collateral loan, a bank uses a borrower's own assets as collateral to secure repayment of the loan when it comes to due. The LTV ratio is an important risk factor used by a bank in qualifying a borrower for a collateral loan. This ratio is also closely monitored by a bank after a loan is being made, because the value of collateral could change constantly. For example, if the collateral value decreases, causing an increase of LTV ratio, then the bank will require the borrower to provide more collateral.

To determine an optimal LTV ratio, a critical issue for a bank is to accurately value loan collateral. However, valuing collateral (and any other financial assets as well) is challenging because the capital market involves both risks and uncertainties. Here we follow Chen and Epstein (2002) and Knight (1921) [1, 2] to differentiate risk from uncertainty or Knightian uncertainty (ambiguity) (some studies (e.g., Chen and Epstein (2002); Izhakian (2012) [3]) use ambiguity to refer to Knightian uncertainty). In particular, risk refers to a condition where an outcome is unknown, but the probabilities of the outcome can be measured. In contrast, uncertainty or ambiguity refers to a condition where not only is the event outcome unknown, but also the probabilities associated with the outcome are unknown in the first place. Simply put, risk can be measured, while uncertainty cannot be measured [1] (a possible reason that we cannot measure uncertainty is the lack of all the information we need to estimate probabilities for an event outcome).

This study makes the first effort to introduce Knightian uncertainty into a general framework in determining collateral value when a bank seeks to set up an optimal LTV ratio. Several studies have examined the determinants of bank LTV ratios and the influence of collateral value by considering various types of risks. For example, following the structural method of Merton (1974) [4], Jokivuolle and Peura (2003) [5] present a model of risky debt in which collateral value is correlated with the possibility of default. Cossin and Hricko (2003) [6] determine the discount rate of the collateral by the structured method. But the models based on the structure method assume the endogeneity of default. In fact, other factors not related to a particular debt, such as the company's liquidity problems, are likely to promote corporate's default. So Cossin et al. (2003) [7] assign an exogenous probability of corporate's default. Moreover Cossin et al. (2003) follow the framework and obtain a discount rate of the collateral which is consistent with a bank's risk tolerance. Li et al. (2006) [8] establish a basic model on the determination of LTV ratios based on a reduced-form approach. But none of these studies have explicitly incorporated inherent Knightian uncertainty into their framework (as pointed by Chen and Epstein (2002), "The Ellsberg Paradox and related evidence have demonstrated that such a distinction (between risk and uncertainty) is behaviorally meaningful" (p 1403). Recently, some economists have invoked Knightian uncertainty to explain the investor behaviors in times of financial crisis (Dizikes, 2010)).

To appropriately value bank loan collateral, following Chen and Epstein (2002) [1], we consider both the risk and Knightian uncertainty in the financial market. We use a set of probability measures to build a bank's minimum and maximum levels of risk tolerance in an environment with Knightian uncertainty. Under the assumption that the short-term prices of the collateral follow a geometric Brownian motion, we study a borrower's default probability impacted by Knightian uncertainty parameter. Applying BSDEs, we get the explicit solutions of the models about a bank's minimum and maximum levels of risk tolerance. Applying the explicit solutions, we build models of the LTV ratios and obtain an interval solution for the optimal loan-to-value ratios. Finally, our numerical analysis is consistent with the interval solution derived from the model.

The remainder of the paper is organized as follows. In Section 2, we first state the assumptions and the definitions of bank's maximum and minimum risk preference. Then we build the models about the LTV ratios and get the explicit solutions of the models. Applying the explicit solutions, we obtain an interval solution for the optimal LTV ratio. In Section 3, we make numerical analysis of the LTV ratio models. Section 4 draws the concluding remarks.

2. Models

Give a filtered probability space (Q, F, [{[F.sub.t]}.sub.0 [less than or equal to] t [less than or equal to] T], P), where the filtration [{[F.sub.t]}.sub.0 [less than or equal to] t [less than or equal to] T] is the [sigma]-algebra generated by the Brownian motion [{[W.sub.t]}.sub.0 [less than or equal to] t [less than or equal to] T]. Suppose that there are two tradable assets in the market. One is a risk-free bond with an interest rate r, and the other is goods used by a borrower as a loan collateral. Their price processes satisfy the following SDEs, where the parameters of r, [mu], [sigma], s are constants, respectively, and T is contractual maturity.

[dP.sub.t] = [P.sub.t]rdt, [P.sub.0] = 1, (1)

[dS.sub.t] = [S.sub.t] ([mu]dt + [sigma][dW.sub.t]), [S.sub.0] = s, 0 [less than or equal to] t [less than or equal to] T. (2)

At the initial time [T.sub.0], the borrower will give [a.sub.0] units of the goods with [S.sub.0] price to a bank in order to apply to a loan. The bank will give a [omega] ratio of loan amount for each unit of the collateral.

The model assumptions are listed as follows:

(1) The loan interest rate is a constant R. The loan principal and interest at time t is [v.sub.t] = [v.sub.0][e.sup.Rt], where [v.sub.0] is the underlying asset.

(2) There is a storing cost of loan collateral during the loan period. The bank will hold the cost credited to the loan interest rate.

(3) The loan contract matures at time T, the frequency of covering short positions is M, and the time interval of covering short positions is [tau]([tau] x M = T). The trigger level of covering short positions is zero; that is, as long as the loanable value of the collateral (market value of the goods x LTV ratio) is less than the sum of loan principal and interest, the borrower will receive a margin call to restore balance.

(4) We assume the default probability of the loan is [Q.sub.0]]. It is exogenously given by the bank.

In the beginning of the m period, where m = 1,2, ..., M, given the interest rate of R, the loan principal and interest is [v.sub.0][e.sup.R[tau](m-1)]. The unit number of the collateral is [a.sub.m-1], with the market value for each unit of [S.sub.m-1]. Thus [v.sub.0][e.sup.R[tau](m-1)] = [omega][a.sub.m-1][S.sub.m-1].

When the loan contract continues to the end of the m period, the following three cases will appear:

(1) When [v.sub.0][e.sup.R[tau]m] = [omega][a.sub.m-1][S.sub.m-1], the loan principal and interest is equal to the loanable value of the collateral at the end of the m period; the borrower does nothing and the contract simply continues.

(2) When [v.sub.0][e.sup.R[tau]m] < [omega][a.sub.m-1][S.sub.m-1], the borrower can reduce some collateral to a level that [v.sub.0][e.sup.R[tau]m] = [omega][a.sub.m-1][S.sub.m-1]; the contract then continues.

(3) When [v.sub.0][e.sup.R[tau]m] > [omega][a.sub.m-1][S.sub.m-1], the borrower must add more collateral so that [v.sub.0][e.sup.R[tau]m] = [omega][a.sub.m-1][S.sub.m-1]; the contract will then continue. Otherwise, the borrower defaults and the contract will be terminated and the liquidation begins. In this case, the bank suffers the loss of [v.sub.0][e.sup.R[tau]m] - [a.sub.m-1][S.sub.m].

2.1. Knightian Uncertainty and Bank's Maximum and Minimum Risk Preference. To consider the financial market with Knightian uncertainty, we introduce a feasibly controllable set: [THETA] = {[([[theta].sub.t]).sub.0 [less than or equal to] t [less than or equal to] T] | [absolute value of [[theta].sub.t]] [less than or equal to] k, a.e. t [member of] [0, T]}, where the constant k is nonnegative. Chen and Epstein (2002) call [THETA] k-ignorance. From SDE (1) and (2), we get

[S.sub.t] = s exp {(r - [1/2] [[sigma].sup.2]) t + [sigma][W.sub.t.sup.Q]}, 0 [less than or equal to] t [less than or equal to] T, (3)

where [W.sub.t.sup.Q] = [[sigma].sup.-1]([mu] - r)t + [W.sub.t]. Let [mathematical expression not reproducible].From the Girsanov theorem, we get that the measure Q is equivalent to the measure P. With the measure Q, the process [{[W.sub.t.sup.Q]}.sub.0 [less than or equal to] t [less than or equal to] T] is a Brownian motion. The set [[PI].sup.[theta]] of equivalent martingale measures is constituted from the set [THETA]. [mathematical expression not reproducible]. The Knightian uncertainty of the financial market is characterized by the set [[PI].sup.[theta]]. Because the bank does not know which probability measure of [[PI].sup.[theta]] should be used to calculate the probability of loss, to be conservative, it will calculate the maximum and minimum probability of loss. That is, for any measurable event A, define

[mathematical expression not reproducible]. (4)

Clearly, when k = 0, there exists a unique equivalent martingale measure Q; thus [P.sub.min](A) = [P.sub.max](A) = P(A). This indicates that the financial market does not involve the Knightian uncertainty but only risk. All of the previous studies have only considered the situation when k = 0, while our study takes the first look when k [greater than or equal to] 0 whereas both the risks and Knightian uncertainty in the financial market are being considered.

2.2. Determining LTV Ratio. We build the maximum and minimum risk preferences for a bank to determine the optimal LTV ratio. At the beginning of the m period, [v.sub.0][e.sup.R[tau](m-1)] = [omega][a.sub.m-1][S.sub.m-1]; thus [a.sub.m-1] = [v.sub.0][e.sup.R[tau](m- 1)]/[omega][S.sub.m-1]. When the borrower defaults, the loss suffered by the bank is [loss.sub.m]. We have

[loss.sub.m] = [e.sup.-rm[tau]] ([v.sub.0][e.sup.R[tau]m] - [a.sub.m-1] [S.sub.m]) = [e.sup.-rm[tau]] ([v.sub.0][e.sup.R[tau]m] - [[v.sub.0][e.sup.R[tau](m - 1)]/[omega]] [[S.sub.m]/[S.sub.m-1]]). (5)

Let L be the maximum loss that the bank is willing to bear, and let L be the function of the underlying asset [v.sub.0]. For simplicity, we let L = l[v.sub.0], where l denotes the degree of loan loss determined by the bank. Then we can calculate the probability of [loss.sub.m] not less than L = l[v.sub.0] in the m-th time interval. We can also get a bank's minimum and maximum risk preferences in the m-th time interval. Here [loss.sub.m] is caused by the uncertainty of collateral prices at the m period.

[mathematical expression not reproducible]. (6)

Next, we will get the explicit solutions of the models. First we give several important lemmas. Let

[L.sup.2](0, T) := {X : [{[X.sub.t]}.sub.0 [less than or equal to] t [less than or equal to] T] be [{[F.sub.t]}.sub.0 [less than or equal to] t [less than or equal to] T] adapted process and [[parallel]X[parallel].sup.2] = E [[integral].sup.T.sub.0] [[absolute value of [X.sub.s]].sup.2]ds < [infinity]},

[L.sup.2]([OMEGA], F, P) := {[xi] : [xi] be [F.sub.T] adapted process and E[[absolute value of [xi]].sup.2] < [infinity]}.

Lemma 1. For any [xi] [member of] [L.sup.2]([OMEGA], F, P), [([[theta].sub.s]).sub.0 [less than or equal to] s [less than or equal to] T] [member of] [THETA], consider the following BSDE:

-d[y.sup.[theta].sub.t] = - (r[y.sup.[theta].sub.t] + [z.sup.[theta].sub.t] x [[theta].sub.t])dt - [z.sup.[theta].sub.t][W.sup.Q.sub.t], [y.sup.[theta].sub.T] = [xi]. (7)

There exist [([[theta].sup.1.sub.t]).sub.0 [less than or equal to] t [less than or equal to] T] ] and [([[theta].sup.2.sub.t]).sub.0 [less than or equal to] s [less than or equal to] T]] in the set [THETA] which satisfy

[mathematical expression not reproducible]. (8)

Proof. From Pardoux and Peng (1990) [9], we know that there exists unique adapted solution [([y.sup.[theta].sub.t], [z.sup.[theta].sub.t]).sub.0 [less than or equal to] t [less than or equal to] T]] [member of] [L.sup.2](0, T) x [L.sup.2](0, T) for the BSDE (7), and [mathematical expression not reproducible]. From El Karoui et al. (1997) [10], we know that

[mathematical expression not reproducible], (9)

[mathematical expression not reproducible], (10)

and [mathematical expression not reproducible] is the solution of the following BSDE.

[mathematical expression not reproducible]. (11)

Similarly, we can get the formula

[mathematical expression not reproducible], (12)

[mathematical expression not reproducible], (13)

and [mathematical expression not reproducible] is the solution of the following BSDE.

[mathematical expression not reproducible]. (14)

Lemma 2. Let [xi] be the indicator function [mathematical expression not reproducible] and assume that the coefficient of diffusion in the SDE (2) a is positive; then [([[theta].sup.1.sub.t]).sub.0 [less than or equal to] t [less than or equal to] T]] [equivalent to] k, [([[theta].sup.2.sub.t]).sub.0 [less than or equal to] t [less than or equal to] T]] [equivalent to] -k. Thus

[mathematical expression not reproducible], (15)

where [mathematical expression not reproducible].

Proof. Let [xi] be the indicator function [mathematical expression not reproducible], from formulas (9) and (12), we get

[mathematical expression not reproducible]. (16)

From Pardoux and Peng (1994) [11], we get the formula

[mathematical expression not reproducible]. (17)

And [mathematical expression not reproducible] are the partial derivatives of [mathematical expression not reproducible] about the tradable goods' price.

We know that the indicator function [mathematical expression not reproducible] is a decreasing function about goods' price. According to the comparison theorems of SDE and BSDE, we get that [mathematical expression not reproducible] are negative. Thus the processes [mathematical expression not reproducible] are all negative. From formulas (10) and (13), we get the conclusion obviously.

Theorem 3.

[mathematical expression not reproducible], (18)

where D([omega]) = ln [omega]([e.sup.R[tau]] - [le.sup.rm[tau]-R(m-1)[tau]]).

Proof. From Lemma 2, we get [mathematical expression not reproducible].

For [Q.sup.(k)] in [[PI].sup.[theta]], we define the process [mathematical expression not reproducible]. Thus the process [mathematical expression not reproducible] is a Brownian motion about

the probability measure [Q.sup.(k)] . From SDE (3), we get

[mathematical expression not reproducible], (19)

where D([omega]) = ln [omega]([e.sup.R[tau]] - l[e.sup.rm[tau]-R(m-1)[tau]]).

Similarly we can get [P.sub.min]([loss.sub.m] [greater than or equal to] L) = N((D([omega]-r + k[sigma] - (1/2)[[sigma].sup.2])[tau])/[sigma][square root of [tau]]), which completes the proof.

If we make a simple assumption that the probability of loan default is exogenous and independent of the particular loan, the bank is concerned with the probability of two events occurring simultaneously. One event is [loss.sub.m] not less than L, and the other event is the borrower defaults. We assume that the loan's default probability per year is [Q.sub.0]] with a uniform distribution. Therefore, the default probability of the loan in the m-th period is simply [tau][Q.sub.0]. We then get the joint probability of these two events under the bank's maximum and minimum risk preference as follows:

[mathematical expression not reproducible]. (20)

Theorem 4. One assumes that the loan's default probability is exogenous and independent of the particular loan. One also assumes that the loan's default probability per year is [Q.sub.0]] with a uniform distribution. During the entire loan period, the probabilities that the bank's losses are not less than L are then estimated as follows:

[mathematical expression not reproducible], (21)

[mathematical expression not reproducible]. (22)

Proof. We know that the borrower can only default once and the probability that the borrower does not default before the m-1 periods is [(1 - [tau][Q.sub.0]).sup.m-1] . Thus, with the bank's and minimum risk preference, the probabilities that the bank losses are not less than L in the m-th period are estimated as [(1 - T[Q.sub.0]).sup.m-1] [P.sub.max](m) and [(1 - [tau][Q.sub.0]).sup.m-1][P.sub.min](m), respectively. During the entire loan period, the probabilities that the bank's losses are not less than L are then estimated as follows:

[mathematical expression not reproducible]. (23)

The proof is completed.

Remark 5. Note that both [P.sub.min](loss [greater than or equal to] L) and [P.sub.max](loss [greater than or equal to] L) reflect a bank's minimum and maximum risk preference for the loan collateral. Given the probabilities of [P.sub.min](loss [greater than or equal to] L) and [P.sub.max](loss [greater than or equal to] L) and the loan's default probability [Q.sub.0]], mark to market frequency M, loan time T, loan interest rates R, and other parameters, we can solve for the numerical solution of LTV ratios or [omega].

3. A Numerical Analysis

We assume that the collateral is copper in the futures market, and its price equation is driven by a Brownian motion, with [sigma] = 0.4118, [mu] = 0.255, and P = 0.1358. Following Li et al. (2007) [12], we assume the following parameters in the numerical analysis: [Q.sub.0] = 0.9 (loan default probability per year), r = 0.03 (interest rate), R = 0.15 (loan interest rate), M = 90 (mark to market frequency), T = 90 days (loan time), and k (Knightian uncertainty parameter) is in (0,1).

We compute the results of formulas (22) and (21) using the Maple software with the minimum and maximum loss probabilities. From formula (22), we get [omega] = 0.57. From formula (21), we get [omega] = 0.87. Thus, after considering the Knightian uncertainty, we get the optimal LTV ratio with an interval of (0.57, 0.87). In the Knightian uncertainty-neutral environment, the LTV ratio [omega] = 0.70274, which is in this interval.

4. Conclusion

We can find that LTV ratios in Knight uncertainty environment are reduced for risk-averse banks, the risk taken is better controlled, and the risk reduction is achieved. Nowadays, because the SMEs encounter the survival bottlenecks, the state has developed a series of measures to ensure the source of funds for SMEs, one of which is to reduce the difficulty of bank loans. Thus the banking institutions were forced to become risk-loving participants. The models of this paper also provide a more reasonable quantitative analysis so as to achieve the tripartite win-win situation among the SMEs and banking institutions and logistics enterprises.

In this paper, we provide a general framework to determine a bank's optimal LTV ratios in the financial market with Knightian uncertainty. In the future, we would extend our models by considering such impacting factors as liquidation delay, liquidity risk, nonzero trigger level, and so on.

https://doi.org/10.1155/2017/1031247

Conflicts of Interest

The authors declare that they have no conflicts of interest.

Acknowledgments

This research was supported by Chinese National Natural Science Foundation (no. 11301298; no. 11671229), Natural Science Foundation of Shandong Province (no. ZR2016GB05), and Shandong Province Social Science Planning Project Foundation (no. 14CJRJ03). The authors are grateful to Professor Ting Zhang for his help.

References

[1] Z. Chen and L. Epstein, "Ambiguity, risk, and asset returns in continuous time," Econometrica, vol. 70, no. 4, pp. 1403-1443, 2002.

[2] EH. Knight, Risk, Uncertainty, and Profit, Hart, Schaffner Marx; Houghton Mifflin Company, Boston, Mass, USA, 1921.

[3] Y. Izhakian, "Capital asset pricing under ambiguity, NYU working paper, 2012".

[4] R. Merton, "On the pricing of corporate debt: the risk structure of interest rates," The Journal of Finance, vol. 29, no. 2, pp. 449-470, 1974.

[5] E. Jokivuolle and S. Peura, "Incorporating collateral value uncertainty in loss given default estimates and loan-to-value ratios," European Financial Management, vol. 9, no. 3, pp. 299-314, 2003.

[6] D. Cossin and T. Hricko, "A structural analysis of credit risk with risky collateral: a methodology for haircut determination," Economic Notes, vol. 32, no. 2, pp. 243-282, 2003.

[7] D. Cossin, Z. Huang, and D. Aunon-Nerin, A Framework for Collateral Risk Control Determination, European central bank working paper, series 1, 2003,1-47.

[8] Y. Li, Y. Xu, G. Feng, and F. Wang, "Research on Loan-to-value Ratios of Standard Inventory Financing," Operations Research and Management Science, vol. 15, pp. 78-82, 2006.

[9] E. Pardoux and S. Peng, "Adapted solution of a backward stochastic differential equation," Systems and Control Letters, vol. 14, no. 1, pp. 55-61, 1990.

[10] N. El Karoui, S. Peng, and M. C. Quenez, "Backward stochastic differential equations in finance," Mathematical Finance, vol. 7, no. 1, pp. 1-71, 1997.

[11] E. Pardoux and S. Peng, Backward Doubly Stochastic Differential Equation and Quasi-Linear PDEs, Lecture Notes in CIS, vol. 176, Springer-verlag, 1994.

[12] Y. Li, G. Feng, and Y. Xu, "Research on loan-to-value ratio of inventory financing under randomly-fluctuant price," System Engineering Theory and Practice, vol. 27, no. 12, pp. 42-49, 2007.

Hui Zhang, (1) Wenyu Meng, (2) Xiaojie Wang, (1) and Jianwei Zhang (3)

(1) School of Mathematics and Quantitative Economics, Shandong University of Finance and Economics, Jinan 250014, China

(2) School of Finance, Shandong University of Finance and Economics, Jinan 250014, China

(3) Shandong Police College, Jinan 250014, China

Correspondence should be addressed to Hui Zhang; drzhanghui@163.com

Received 16 March 2017; Accepted 8 May 2017; Published 4 June 2017

Academic Editor: Konstantinos Karamanos

Printer friendly Cite/link Email Feedback | |

Title Annotation: | Research Article |
---|---|

Author: | Zhang, Hui; Meng, Wenyu; Wang, Xiaojie; Zhang, Jianwei |

Publication: | Discrete Dynamics in Nature and Society |

Date: | Jan 1, 2017 |

Words: | 3976 |

Previous Article: | Comment on "Unilateral Global Bifurcation from Intervals for Fourth-Order Problems and Its Applications". |

Next Article: | Qualitative Stability Analysis of an Obesity Epidemic Model with Social Contagion. |

Topics: |