Quadratic -BSDEs for Bond Pricing with Endogenous Short-Rate Feedback
Abstract
We study robust bond valuation with endogenous short-rate feedback under volatility uncertainty. Within the -expectation framework, the dependence of the short rate on the bond price yields a nonlinear fixed-point problem, represented by a quadratic -BSDE for the logarithmic price. Under suitable assumptions, we establish existence, uniqueness, comparison, and stability for bounded finite-horizon solutions. An additional strict monotonicity condition yields a unique bounded infinite-horizon solution and exponential convergence of finite-horizon approximations on compact time intervals. We apply these results to inverse short-rate design, constructing discount-rate coefficients that reproduce admissible smooth bond-price targets at a fixed maturity. For long maturities, we construct feedback rules under which the compensated logarithmic price converges exponentially to a prescribed bounded state-dependent profile, while the asymptotic yield equals a specified target.
1 Introduction
The standard short-rate approach values a zero-coupon bond by discounting its terminal payoff using a short-rate process specified independently of the price being computed. A different problem arises when interest-rate decisions respond to bond-market valuations. Bond prices then depend on the future path of interest rates, while the interest-rate rule itself depends on bond prices. This circular dependence turns valuation into a self-consistency problem. When the law of the underlying economic state is also uncertain, the problem combines two distinct nonlinearities: endogenous price–rate feedback and valuation under multiple volatility scenarios. This paper develops a -BSDE framework for analyzing their interaction.
Endogenous feedback between financial valuations and economic decisions is a recurring theme in economic theory. Kiyotaki and Moore, (1997) show how the interaction between asset prices and collateral constraints can amplify and propagate shocks. In a continuous-time macroeconomic model with a financial sector, Brunnermeier and Sannikov, (2014) analyze nonlinear feedback between financial conditions, asset valuations, and endogenous risk. Focusing more directly on interest-rate decisions, Bernanke and Woodford, (1997) examine the interaction between monetary policy and private-sector forecasts, emphasizing the difficulties that arise when the information used to guide policy responds to the policy rule itself. These contributions motivate a joint analysis of financial variables and the rules that respond to them. Our model isolates a reduced-form price–rate feedback mechanism; it does not derive that mechanism from a general equilibrium or a welfare-maximization problem.
The treatment of uncertainty is likewise important. Specifying stochastic volatility under a single probability measure allows volatility to fluctuate, but still prescribes its probabilistic law. The -expectation framework instead accommodates a family of volatility scenarios through a sublinear expectation, with -Brownian motion as the driving noise (Denis et al.,, 2011; Peng,, 2019). The resulting quadratic variation is itself uncertain. This feature permits the discounting rule to distinguish between ordinary time and realized quadratic variation, and it leads naturally to nonlinear valuation equations.
The financial implications of volatility ambiguity have been studied from both hedging and equilibrium perspectives. Epstein and Ji, (2013) develop a continuous-time asset-pricing framework with ambiguity about volatility and examine its implications for hedging and equilibrium asset returns. Vorbrink, (2014) formulates financial markets with volatility uncertainty using -Brownian motion and studies upper and lower arbitrage-free prices. These works clarify why robust valuation need not select a unique price independently of the pricing criterion, preferences, or market structure. In the present paper, uniqueness refers to a self-consistent valuation under a specified conditional sublinear expectation, rather than uniqueness across all possible ambiguity-sensitive pricing rules.
For bond markets, Hölzermann, (2021) constructs a Hull–White model under volatility uncertainty. Hölzermann, (2022) subsequently develops a Heath–Jarrow–Morton framework driven by -Brownian motion. Of particular relevance is Theorem 3.1 of that paper: under suitable regularity assumptions, a drift condition involving a market price of risk and additional market prices of uncertainty yields a sublinear expectation under which discounted bonds are symmetric -martingales, and hence excludes arbitrage in the specified market. The condition also determines the risk-neutral forward-rate drift from the volatility structure. The appearance of several market prices reflects the distinction between ordinary drift and quadratic-variation-dependent drift under volatility uncertainty. This is a sufficient no-arbitrage result; its symmetry condition is stronger than the -martingale property alone.
Further developments include Hölzermann, (2024), who studies the pricing of interest-rate derivatives under volatility uncertainty, and Akhtari and Li, (2024), who establish a -expectation formulation of the Cox–Ingersoll–Ross process and investigate its mathematical properties. Together, these contributions provide models and analytical tools for interest rates under uncertain volatility. Our focus is different: rather than taking the discount-rate coefficients as independent inputs to valuation, we allow them to depend on the logarithmic value being determined. The key questions are whether the resulting implicit pricing equation is well posed and how the feedback rule shapes its solution.
To describe the problem, let be a state process driven by a -dimensional -Brownian motion , and let denote the conditional sublinear expectation used for valuation. Fix a reference maturity and write . We consider generalized discount-rate coefficients and , acting through ordinary time and quadratic covariation, respectively. The self-consistency requirement is
Setting recovers feedback through the ordinary short rate alone. In the general formulation, the are coefficients of the discounting rule; they are not the market prices of uncertainty appearing in the drift condition of Hölzermann, (2022).
The logarithmic transformation turns this fixed-point problem into the quadratic -BSDE
where is a decreasing -martingale. The quadratic term in comes from the logarithmic transformation, whereas records the nonlinear martingale component associated with volatility uncertainty. This representation makes it possible to study self-consistent valuation using estimates and comparison arguments for quadratic backward equations. The valuation problem is formulated for a fixed reference maturity; compatibility across simultaneously traded maturities under one common money-market account is a separate requirement.
The relevant mathematical foundations are provided by the literature on -BSDEs. Hu et al., 2014a (); Hu et al., 2014b () establish the Lipschitz theory and develop comparison, nonlinear Feynman–Kac, and Girsanov results. Hu et al., (2018) study quadratic -BSDEs through discrete approximations and fully nonlinear partial differential equations. For the quadratic and long-horizon questions considered here, several extensions are particularly relevant. Hu and Wang, (2018) develop infinite-horizon and ergodic -BSDEs and their applications. Hu et al., (2022) treat quadratic -BSDEs with convex generators and unbounded terminal conditions. Sun and Wang, (2024) connect ergodic quadratic -BSDEs with homothetic forward performance processes under volatility uncertainty. More recently, Lin et al., (2026) establish well-posedness and comparison results for infinite-horizon quadratic -BSDEs.
The work of Falei Wang and his coauthors provides a central methodological foundation for our analysis. In particular, our treatment of infinite-horizon equations builds on the linearization and finite-horizon approximation methods of Hu and Wang, (2018). Their linearization result is used explicitly in the proof of Lemma 4.2, where it is combined with truncation and -BMO estimates to handle quadratic growth. These methods underpin the well-posedness and horizon-convergence analysis in Section 4, which we apply to self-consistent bond valuation and long-maturity short-rate design.
Building on this theory, the present paper addresses two connected financial questions: the well-posedness of endogenous bond valuation and the inverse design of short-rate feedback for prescribed price targets. We first develop a Markovian analysis suited to the pricing application. With bounded terminal data, appropriate Lipschitz assumptions, and quadratic growth in the martingale integrand, we establish a priori bounds and BMO estimates, together with a comparison principle, stability, and the existence and uniqueness of bounded solutions. We also establish spatial Lipschitz continuity and temporal -Hölder continuity of the associated Markovian value function. Applied to the logarithmic pricing equation, these results yield a unique bounded self-consistent valuation under the stated feedback assumptions.
We next consider infinite-horizon equations under a strict monotonicity condition in the backward variable. The corresponding estimates yield a unique bounded infinite-horizon solution and show that finite-horizon approximations converge exponentially on every compact time interval. Thus the influence of a remote terminal condition can be controlled quantitatively. This horizon-stability result is the analytical link between finite-maturity valuation and the long-maturity design problem.
Finally, we study the inverse question of constructing discount-rate coefficients from a prescribed price target. For a fixed maturity, an admissible smooth function , satisfying , determines coefficients that reproduce . For long maturities, we prescribe a target yield and an admissible smooth bounded profile , and use feedback through the compensated logarithmic price
A linear restoring term , with , together with suitable state-dependent discount-rate coefficients, identifies as the bounded infinite-horizon profile. Under the required regularity and boundedness conditions, the finite-horizon solutions satisfy
Consequently, for each fixed ,
Here is prescribed as a design target, not determined as an ergodic eigenvalue. Likewise, controls convergence as maturity increases; the result is not a statement about the speed of adjustment following a policy intervention in calendar time.
The remainder of the paper is organized as follows. Section 2 recalls the -expectation framework and the BMO tools used in the analysis. Section 3 studies finite-horizon quadratic -BSDEs, and Section 4 develops the infinite-horizon theory and convergence estimates. Section 5 applies these results to bond valuation with endogenous short-rate feedback. Section 6 considers fixed-maturity and long-maturity target design. Section 7 concludes.
2 Preliminaries
2.1 G-expectation
In this subsection, we briefly recall the basic notions and results of -expectation theory. We refer the reader to Denis et al., (2011), Hu et al., 2014a (), and Peng, (2019) for further details. Let be the canonical space of all -valued continuous paths starting from , equipped with the metric
Let denote the canonical process on , that is, for . For each , define
| (2.1) |
Here, denotes the space of all continuous functions for which there exist constants and , depending only on , such that for all . It is immediate that whenever . We therefore set .
Let denote the space of all symmetric matrices. Given a monotone sublinear functional , there exists a sublinear expectation space under which the canonical process is a -Brownian motion satisfying
| (2.2) |
Moreover, there exists a bounded and closed set of symmetric positive semidefinite matrices such that
| (2.3) |
The conditional -expectation is initially defined for by
Here, is represented as , where and for some . The sublinear expectation space is called a -expectation space, while and are referred to as the -expectation and the conditional -expectation, respectively.
For , we define (, respectively) as the completion of (, respectively) with respect to the norm . For each , the conditional sublinear expectation extends continuously to . In this paper, we shall only consider -Brownian motions satisfying the strong ellipticity condition, that is, there exist strictly positive constants such that
| (2.4) |
The conditional -expectation admits a probabilistic representation. For the following representation theorem, see Hu and Peng, (2009); Hu and Peng, (2021). Let be the filtration generated by the canonical process and define .
Theorem 2.1.
The -expectation can be represented by a weakly compact family of probability measures on . More precisely,
| (2.5) |
The family is called a representing set for . Furthermore, for every , the conditional -expectation has the following representation:
| (2.6) |
where
We define the capacity as
| (2.7) |
where . A set is called polar if . We say that a property holds “quasi-surely” if it holds outside a polar set. Let be the set of all bounded -measurable real-valued functions, and define as the completion of with respect to the norm . We define (, respectively) as the completion of (, respectively) with respect to the norm . Denis et al., (2011) proved that
| (2.8) |
Consequently, to establish , it suffices to show that for some .
We now consider several process spaces and norms. For and , define
| (2.9) | ||||
| (2.10) |
with norms
| (2.11) |
Here, is the space of all bounded Lipschitz continuous functions on The completions of with respect to the norms and are denoted by and , respectively. The completion of with respect to the norm is denoted by . The corresponding spaces , , and are defined analogously for . For , the spaces , , and are defined as the -fold product spaces , , and , respectively. Moreover, for , we define , , and .
Following Li and Peng, (2011), for and , we define the integrals and . We next introduce -martingales and recall some of their properties.
Definition 2.1.
A process is called a -martingale if for and for . A process is called a symmetric -martingale if both and are -martingales.
2.2 BMO martingales
This subsection introduces -BMO martingale generators and recalls some of their properties. We refer the reader to Hu et al., (2018) for further details.
Definition 2.2.
Let and We say is a -BMO martingale generator if satisfies
| (2.12) |
where denotes the set of all -stopping times taking values in
We can define a new sublinear expectation and sublinear conditional expectation on a -expectation space using a -BMO martingale generator. For fixed let be a -BMO martingale generator. Then the process
| (2.13) |
is a symmetric -martingale. Define a sublinear expectation and a sublinear conditional expectation as
| (2.14) |
for , respectively. We say (, respectively) is the sublinear expectation (the sublinear conditional expectation, respectively) induced by The sublinear expectations and are equivalent. Moreover, the process
| (2.15) |
is a -Brownian motion under the sublinear expectation . It follows that the quadratic variations of and coincide under and For further details, see Xu et al., (2011); Hu et al., (2018).
Lemma 2.2.
Define a function If is a -BMO martingale generator with for , then there is a constant such that
| (2.16) |
The exponent is called the order of the reverse Hölder inequality for . In addition, if is a decreasing -martingale such that and for , then is a decreasing -martingale under .
Lemma 2.3.
Let be a -BMO martingale generator, let be the order of the reverse Hölder inequality for , and let be a decreasing -martingale such that and for . Then is a decreasing -martingale under .
3 Finite-horizon quadratic -BSDEs
Consider the SDE and -BSDE
| (3.1) | ||||
| (3.2) | ||||
| (3.3) |
for , , Occasionally, we write as respectively, omitting the superscripts
Assumption 3.1.
Assume the functions satisfy the following properties.
- 1.
For , .
- 2.
The functions are continuous in .
- 3.
There is a constant such that
(3.4) for .
The following theorem states the existence, uniqueness and regularity of solutions to the SDE (3.1).
Theorem 3.1.
Proof.
Existence and uniqueness follow from (Peng,, 2019, Theorem 5.1.3). The three inequalities are consequences of Lemma 2.2 and the results in Chapter 5 of Peng, (2019). ∎
We now focus on the BSDE (3.2). We first state the assumptions on the drivers and and the terminal condition .
Assumption 3.2.
Assume the functions satisfy the following properties.
- 1.
For ,
- 2.
The functions and are continuous in .
- 3.
There is a constant such that
(3.8) (3.9) (3.10) for , ,
- 4.
There is a constant such that
(3.11) for .
Definition 3.1.
The remainder of this section studies the main properties and existence of solutions to the BSDE. We first establish uniqueness, a comparison principle, regularity, and stability in Subsection 3.1. We then prove the existence of solutions in Subsection 3.2. The financial application to bond pricing with endogenous short-rate feedback is developed separately in Section 5.
3.1 A priori estimates
Theorem 3.2.
Proof.
For simplicity, assume and We first prove that is a -BMO martingale generator and that the norm is bounded by a constant depending only on , , and , and then show that for every Finally, we prove that the norm is bounded by a constant depending only on , , and
Applying Itô’s formula to for a constant which will be specified later, we have
| (3.13) | ||||
| (3.14) |
where and Since
| (3.15) | ||||
| (3.16) |
Setting ,
| (3.17) | ||||
| (3.18) |
This implies that is a -BMO martingale generator and that is bounded by a constant depending on , , and the norm
In order to show for every , we first observe that for all . Applying Itô’s formula, we have
| (3.19) | ||||
| (3.20) |
which implies that . Since is a solution to (3.2), there exists a constant such that
| (3.21) |
Using (Gao,, 2009, Theorem 2.1) and , we obtain that is bounded by a constant depending on and the norms , .
Finally, we show that the norm is bounded by a constant depending on and We use the linearization argument from the proof of (Hu et al.,, 2018, Proposition 3.5). For let be a Lipschitz function satisfying . Define three processes
| (3.22) | ||||
| (3.23) | ||||
| (3.24) |
Then , and It follows that
| (3.25) |
By (Hu et al.,, 2018, Lemma 3.6), the process is a -BMO martingale generator. Moreover, and belong to . Let be the sublinear expectation induced by . By (2.15), the process is a -Brownian motion under We have
| (3.26) |
Since for all , by Lemma 2.3, the process is a decreasing -martingale under . Let be a solution to the SDE
| (3.27) |
It is easy to show that is bounded. Applying Itô’s formula to , we have
| (3.28) |
Since and are bounded by a constant depending only on , , and , under we have
| (3.29) | ||||
| (3.30) |
for some constant depending only on , , and . Since , is bounded uniformly in . Hence Lemma 2.2 applies with a constant independent of . Applying Hölder’s inequality and Lemma 2.2, using , and letting , we conclude that is bounded by a constant depending only on , , , and . ∎
The following theorem establishes a comparison principle for quadratic -BSDEs.
Theorem 3.3.
Proof.
For simplicity, assume and . First, Theorem 3.2 implies that and are -BMO martingale generators. Let and . For , let be a Lipschitz function satisfying . Define four processes
| (3.33) | ||||
| (3.34) | ||||
| (3.35) | ||||
| (3.36) |
Then , and . It follows that
| (3.37) |
By (Hu et al.,, 2018, Lemma 3.6), is a -BMO martingale generator. Then we have
| (3.38) |
Let be the sublinear expectation induced by . By (2.15), is a -Brownian motion under . Let be a solution to the SDE
| (3.39) |
Applying Itô’s formula to under , we have
| (3.40) | ||||
| (3.41) |
Consider the following -BSDE under :
| (3.42) |
It follows that is a solution. By (Hu et al., 2014a, , Theorem 4.1) and (Hu et al., 2014b, , Theorem 3.2), we obtain
| (3.43) |
Applying to both sides of (3.40) and using Lemma 2.3 and Theorem 3.2, we obtain
| (3.44) | ||||
| (3.45) |
By (3.43), we have
| (3.46) |
Letting , we obtain . ∎
Theorem 3.4.
Proof.
For simplicity, we assume that and . Let and be two bounded solutions to (3.2). Applying Theorem 3.3 yields for ,
| (3.47) |
Hence, for every , the processes and are modifications of each other under . Since both processes have continuous paths -almost surely, they are indistinguishable under . Therefore, for every ,
| (3.48) |
Taking the supremum over , we obtain
| (3.49) |
By (Hu et al.,, 2018, Proposition 3.7) and (3.49), we obtain
| (3.50) |
Thus, in . Finally, we have . Thus, (3.2) has at most one bounded solution. ∎
We now establish regularity and growth estimates with respect to the initial condition .
Proposition 3.5.
Proof.
The proof is similar to that of Theorem 3.3, so we provide only an outline. For simplicity, we assume and . By Theorem 3.2, and are -BMO martingale generators. Let and . As in the proof of Theorem 3.3, for , there exist and such that
| (3.53) |
Let be the sublinear expectation induced by . By (2.15), the process is a -Brownian motion under . Define a process similarly. Then we have
| (3.54) | ||||
| (3.55) | ||||
| (3.56) |
Similarly
| (3.57) |
Then we have
| (3.58) | ||||
| (3.59) |
Letting and using Theorem 3.1, we obtain . Moreover, in the same way, we have . Combining these inequalities gives . Similarly, we obtain . ∎
Remark 3.1.
The order of the reverse Hölder inequality for depends on . Therefore, the constant in Proposition 3.5 also depends on and .
Let be a solution to (3.2) under Assumptions 3.1 and 3.2. Define a function
| (3.60) |
Since , , , , , and are deterministic functions and is a -Brownian motion, it follows that is deterministic. If the process is bounded for all , then by Proposition 3.5, we get
| (3.61) | |||
| (3.62) |
The following result plays a key role in proving the existence of solutions to the -BSDE.
Theorem 3.6.
Proof.
For simplicity, we assume and . We first prove the result for bounded . For and , define the simple random variable
| (3.64) |
Using Proposition 3.5, we have
| (3.65) | ||||
| (3.66) |
Using (3.61) and the same argument, we obtain
| (3.67) |
Then we have . Letting , we obtain for bounded . For general , there exists a sequence of bounded random variables such that . Passing to the limit yields .
∎
Theorem 3.7.
Proof.
For simplicity, we assume and . We use an argument similar to that in the proof of Theorem 3.2. For , let be a Lipschitz function satisfying . Define three processes
| (3.69) | ||||
| (3.70) | ||||
| (3.71) | ||||
| (3.72) |
Then , and . It follows that
| (3.73) | ||||
| (3.74) |
Let be the sublinear expectation induced by . By (2.15), is a -Brownian motion under . We have
| (3.75) |
Since for all , by Lemma 2.3, the process is a decreasing -martingale under . Let be a solution to the SDE
| (3.76) |
Applying Itô’s formula to , we have
| (3.77) |
Since is deterministic, for we have
| (3.78) |
From for and Theorem 3.4, we have . Then by Theorem 3.6, it follows that . It follows that
| (3.79) | ||||
| (3.80) | ||||
| (3.81) | ||||
| (3.82) |
By Proposition 3.5, Theorem 3.1, and the boundedness of , we obtain
| (3.83) | ||||
| (3.84) | ||||
| (3.85) |
where the constant depends on , , , and . Letting yields the desired result.
∎
We now establish a stability result.
Theorem 3.8.
Proof.
For simplicity, assume and . We first show that, for some ,
| (3.90) |
Let . By Theorem 3.7, Theorem 3.2 and (3.61), we have
| (3.91) | ||||
| (3.92) |
where the constant depends only on , , , , and , and is independent of the sequence index. By the Arzelà–Ascoli theorem, there exist a function and a subsequence such that converges to uniformly on every compact subset of . For notational simplicity, we relabel as . It follows that
| (3.93) | ||||
| (3.94) | ||||
| (3.95) | ||||
| (3.96) |
where , and the constant depends on , , , and . Sending , we have
| (3.97) |
For , define the process .
Next, we show that, for every ,
| (3.98) |
Define the triple and . Applying Itô’s formula to , we have
| (3.99) | ||||
| (3.100) |
By (Gao,, 2009, Theorem 2.1), we have
| (3.101) | ||||
| (3.102) | ||||
| (3.103) |
where the constant depends on , , , . By Theorem 3.2, for every , , and are bounded uniformly. From (3.97), it follows that . We then define the process
| (3.104) |
Next, we establish the following convergence result:
| (3.105) |
A direct calculation gives
| (3.106) | ||||
| (3.107) | ||||
| (3.108) |
We have , and this random variable converges to quasi-surely. Also where the constant depends on and . Applying (Hu and Zhou,, 2019, Theorem 3.2) and Hölder’s inequality, we have
| (3.109) |
We now show that forms a Cauchy sequence in for every and . Since is a solution, we have
| (3.110) |
Thus, forms a Cauchy sequence in . Denote the limit by . Finally, we show that is a decreasing -martingale. For each , we have
| (3.111) |
Thus, the triple is a solution. ∎
An argument similar to that used in the proof of Theorem 3.8 yields the following corollary.
Corollary 3.9.
Suppose Assumption 3.1 holds and, for each , let the triple be a solution to the following -BSDE:
| (3.112) | ||||
| (3.113) |
where satisfy Assumption 3.2 uniformly. If , and tends to zero as both indices tend to infinity and is bounded uniformly, then there exists a solution to the following equation
| (3.114) | ||||
| (3.115) |
3.2 Existence of solutions
We now state the main result of this section.
Theorem 3.10.
Proof.
For simplicity, assume and . Uniqueness follows from Theorem 3.4: if a triple is a solution to (3.2) and is bounded, then it is the unique solution in this class.
We now prove existence. For each , define the following two functions:
| (3.116) |
Define the nonincreasing Lipschitz function by
| (3.117) |
For each , let Then is bounded and Lipschitz continuous in its argument and nondecreasing with respect to . Define
| (3.118) |
It follows that , and , as . Moreover, we have , and satisfy Assumption 3.2 uniformly. A direct calculation also gives, for every ,
| (3.119) |
Then by (Hu et al., 2014a, , Theorem 4.1), for fixed , the following -BSDE has a unique solution
| (3.120) |
It follows that
| (3.121) |
For fixed and , let . By (Hu et al., 2014b, , Theorem 3.6), the random variable is increasing in for . Thus, we define the process by
| (3.122) |
Applying Itô’s formula to , we have
| (3.123) | ||||
| (3.124) | ||||
| (3.125) | ||||
| (3.126) |
Taking the conditional sublinear expectation on both sides, we obtain
| (3.127) | ||||
| (3.128) |
By (Kim and Park,, 2024, Theorem 2.4), we have for some
| (3.129) | ||||
| (3.130) |
Then by (Hu and Zhou,, 2019, Theorem 3.2) and (3.121), we have
| (3.131) |
We now show that, for each , is a -BMO martingale generator. Applying Itô’s formula to , we obtain
| (3.132) | ||||
| (3.133) |
Since , the norm is bounded by a constant independent of . By an argument similar to that used in the proof of Theorem 3.2, is uniformly bounded. An argument similar to that used in the proof of Theorem 3.8 yields the following convergence results:
| (3.134) | ||||
| (3.135) | ||||
| (3.136) |
These convergence results allow us to define the processes and by
| (3.137) | ||||
| (3.138) |
It follows that the triple is a solution to the following -BSDE:
| (3.139) |
By (3.121), the constructed process is bounded. Since and satisfy Assumption 3.2, for fixed we apply Theorem 3.8 to obtain a triple solving the following -BSDE:
| (3.140) |
Theorem 3.11.
Proof.
The result follows from (Fleming and Soner,, 2006, Lemma 6.2), Dini’s theorem, and (Hu et al., 2014b, , Theorem 4.5). ∎
4 Infinite-horizon quadratic -BSDEs
Fix and coefficient functions and Consider the following forward SDE and -BSDE:
| (4.1) | ||||
| (4.2) | ||||
| (4.3) |
When no confusion can arise, we write as respectively, omitting the superscripts
Assumption 4.1.
Assume that satisfy the following conditions.
- 1.
For , .
- 2.
The functions are continuous in .
- 3.
There exists a constant such that
(4.4) for .
Assumption 4.2.
Assume that satisfy the following conditions.
- 1.
For , .
- 2.
The functions are continuous in .
- 3.
There exists a constant such that
(4.5) (4.6) for , , ,
- 4.
There exists a constant such that
(4.7) for , .
- 5.
There exists a constant such that
(4.8) for , , .
Definition 4.1.
Theorem 4.1.
Proof.
For simplicity, assume and . We first prove uniqueness. Suppose and are solutions to (4.2). Since the processes are bounded by a common constant , Theorem 3.2 implies that, for every ,
| (4.9) |
Let and, for , define . Define the following three processes:
| (4.10) | ||||
| (4.11) | ||||
| (4.12) |
A direct calculation gives , , and . By Lemma 4.2 below, is a -BMO martingale generator on every finite time interval. Let solve the SDE . An argument similar to that used in the proof of Theorem 3.3 yields
| (4.13) | ||||
| (4.14) |
Using , we obtain
| (4.15) | ||||
| (4.16) | ||||
| (4.17) |
Letting , we obtain
| (4.18) |
Letting gives . Interchanging the two solutions gives . Theorem 3.4 then yields uniqueness of the full solution triple.
We now prove existence. For fixed , let the triple be the solution to the following finite-horizon -BSDE:
| (4.19) |
Here, . As in the uniqueness argument, there exist three processes and such that
| (4.20) |
Let be the sublinear expectation induced by . By (2.15), the process is a -Brownian motion under . It follows that
| (4.21) | ||||
| (4.22) |
We extend the triple to by setting
| (4.23) |
For , set . As in the uniqueness argument, there exist three processes and such that
| (4.24) | ||||
| (4.25) | ||||
| (4.26) |
Let be the sublinear expectation induced by . By (2.15), the process is a -Brownian motion under . It follows that
| (4.27) | ||||
| (4.28) |
Interchanging the two solutions in the same argument yields
| (4.29) |
Thus
| (4.30) |
Therefore, for , we obtain
| (4.31) |
For fixed , the restriction of to this interval solves the following -BSDE:
| (4.32) |
Since there exists a function such that , Corollary 3.9 yields the existence of a solution to (4.2). ∎
Lemma 4.2.
The process defined by (4.11) is a -BMO martingale generator.
Proof.
For simplicity, assume . Since , it remains to show that, for every , . For each , define the process
| (4.33) |
By Lemma 3.1 of Hu and Wang, (2018), for every , . Moreover,
| (4.34) | ||||
| (4.35) | ||||
| (4.36) |
Using together with Proposition 3.8 of Li and Peng, (2011), we conclude that . From , it follows that
| (4.37) |
This completes the proof. ∎
We now relate the finite- and infinite-horizon equations by showing that finite-horizon solutions converge to the infinite-horizon solution as the terminal horizon tends to infinity. Consider the finite-horizon -BSDE
| (4.38) | ||||
| (4.39) |
and the infinite-horizon -BSDE
| (4.40) | ||||
| (4.41) |
Theorem 4.3.
Proof.
This estimate follows from (4.30) when is a positive integer satisfying . The extension to arbitrary real follows by repeating the same estimates with that terminal horizon. ∎
5 Bond pricing with endogenous short-rate feedback
In this section, we study bond pricing with endogenous short-rate feedback under -expectation. We formulate the financial market on a -expectation space , where the canonical process is a -dimensional -Brownian motion under . The market consists of the money-market account
where the short-rate processes and are specified below, together with zero-coupon bonds for all maturities satisfying .
To rule out arbitrage opportunities, following Hölzermann, (2022, Theorem 3.1), we assume that there exists a process such that the discounted bond prices
are symmetric -martingales under the sublinear expectation . We recall the definition of in (2.14). The process
| (5.1) |
is a -Brownian motion under . We refer to as the risk-neutral sublinear expectation. Since , the money-market account can equivalently be written as
Accordingly, the time- price of a zero-coupon bond with maturity is given by
| (5.2) |
For notational convenience, we write the logarithmic bond price as . When the maturity is clear from the context, we suppress the superscript and write .
We now introduce endogenous short-rate feedback. Specifically, we assume that the short rate is a function of time, the economic state, and the bond price. Let be an economic state process representing macroeconomic factors such as inflation and unemployment. We assume that solves the -SDE (3.1) with and . The short-rate processes are then specified by
Substituting these feedback specifications into (5.2) yields
| (5.3) |
Unlike the standard bond-pricing formula, (5.3) is implicit: the short rate depends on the bond price through . The existence and uniqueness of a process satisfying (5.3) therefore require a separate argument.
Assumption 5.1.
Suppose that there exist positive constants and such that
| (5.4) | ||||
| (5.5) |
for all , , and .
We seek to establish the existence and uniqueness of a process satisfying (5.3). To this end, we transform the implicit bond-pricing equation (5.3) into the following quadratic -BSDE:
| (5.6) | ||||
The following theorem establishes the well-posedness of this -BSDE and, consequently, of the endogenous bond-pricing equation (5.3).
Theorem 5.1.
Proof.
For simplicity, assume . The coefficients and induce generators satisfying the assumptions of Theorem 3.10, which yields a unique solution to (5.6).
We first verify that satisfies (5.3). Applying Itô’s formula to , we obtain
| (5.7) |
Therefore, satisfies (5.3). To prove uniqueness for the implicit pricing equation, let be another bounded process satisfying (5.3). Then, for each ,
| (5.8) |
By the -martingale representation theorem, there exist and a decreasing -martingale such that
| (5.9) |
Applying Itô’s formula again yields
| (5.10) | ||||
| (5.11) | ||||
| (5.12) |
Define
| (5.13) | ||||
| (5.14) |
The process
is positive and bounded. Hence, by Lemma 3.4 of Hu et al., 2014a (), is a decreasing -martingale, and is a solution to (5.6). Uniqueness in Theorem 3.10 yields . Consequently, is the unique bounded process in satisfying (5.3). ∎
6 Policy short-rate design for target bond prices
6.1 Fixed-maturity bonds
We consider the inverse problem of choosing a policy short rate to reproduce a prescribed bond-price target. Let be a prescribed function, and suppose that the monetary authority seeks to attain the target bond price
for a zero-coupon bond with maturity . This formulation allows the target to depend on macroeconomic conditions, such as inflation and unemployment, through the state process .
For a fixed maturity, the construction follows by matching the dynamics of the target price. We present this result as preparation for the long-maturity problem in Subsection 6.2. The proposition below follows directly from Itô’s formula, so we omit its proof.
Proposition 6.1.
Let satisfy . Define
and
| (6.1) |
for . Suppose Assumption 3.1 holds and that and are bounded; that is, there exists a constant such that
for all . Then the target satisfies
| (6.2) |
6.2 Long-maturity bonds
The preceding subsection considers an inverse problem for a single, fixed maturity. We now study the corresponding long-maturity problem. Suppose that the monetary authority specifies a target long-term yield . To remove the linear decay in the logarithmic price associated with this target, we define the compensated logarithmic bond price
| (6.3) |
Thus,
represents the bond price after removing the prescribed exponential decay with respect to maturity.
The feedback coefficients now depend on , rather than on the uncompensated logarithmic bond price. The resulting bond-price process is therefore required to satisfy
| (6.4) |
By the positive homogeneity of , this equality is equivalent to
| (6.5) |
Let be the target profile for the compensated logarithmic bond price. The authority seeks feedback coefficients such that, for each fixed ,
| (6.6) |
The first limit specifies the state-dependent profile that remains after removing the prescribed long-maturity decay. It also implies the long-term yield identity in the second limit.
Theorem 6.2.
Fix , a function , and a constant . Define
and
Suppose that Assumption 4.1 holds and that there exist positive constants such that
| (6.7) |
for all and . Then the following statements hold.
- (i)
The finite-horizon quadratic -BSDE
(6.8) (6.9) admits a unique solution with bounded.
- (ii)
The process is the unique bounded process in satisfying (6.5).
- (iii)
The infinite-horizon quadratic -BSDE
(6.10) (6.11) admits a unique solution with bounded. Moreover, this solution satisfies .
- (iv)
The parameter in the feedback term controls the rate at which the finite-maturity solution approaches the prescribed long-maturity profile. When the compensated logarithmic bond price exceeds its target, the term increases the specified policy rate; when the price falls below its target, the term decreases the rate. As shown by (6.12), convergence is exponential with rate as maturity increases. Thus, selecting determines the rate in the horizon-convergence estimate. This estimate concerns convergence with respect to maturity, not calendar-time adjustment after a policy intervention.
7 Conclusion
This paper studies bond valuation when the discounting rule depends on the bond price itself and the underlying state evolves under volatility uncertainty. The feedback leads to an implicit pricing problem, rather than the evaluation of a payoff under an independently specified short rate. Representing the logarithmic bond price as the solution of a quadratic -BSDE allows us to analyze this self-consistency problem through existence, uniqueness, comparison, and stability results.
The finite-horizon analysis yields a unique bounded valuation under suitable assumptions, together with BMO estimates and Markovian regularity. Under an additional strict monotonicity condition, the infinite-horizon equation admits a unique bounded solution, and finite-horizon approximations converge exponentially on compact time intervals. These results address both the well-posedness of the feedback problem at a fixed maturity and its behavior as maturity tends to infinity.
The inverse constructions determine discount-rate coefficients from a prescribed price profile. At a fixed maturity, admissible smooth targets are reproduced by matching their -Itô dynamics. For long maturities, a prescribed yield and an admissible bounded state-dependent profile are combined with linear restoring feedback so that the compensated logarithmic price approaches the target exponentially. The feedback coefficient determines the rate of convergence with respect to maturity, while the target yield determines the asymptotic exponential decay of the bond price. These results concern implementability within the specified model, not welfare-optimal policy.
Several questions remain outside the scope of these valuation results. The sublinear pricing expectation is specified as part of the model. Uniqueness of its pricing fixed point does not by itself establish a unique equilibrium price or the symmetric-martingale conditions needed to rule out arbitrage in a traded bond market. Extending feedback from a reference bond to several maturities under a common short-rate process is therefore a natural next step. Other directions include restrictions on admissible rate coefficients, numerical approximation of the pricing fixed point, and statistical specification of the volatility-uncertainty set. The present analysis provides a basis for these extensions by connecting robust valuation, horizon stability, and target-oriented short-rate design within a common backward-equation framework.
Acknowledgements.
Hyungbin Park was supported by the National Research Foundation of
Korea (NRF) grants funded by the Ministry of Science and ICT (2021R1C1C1011675, 2022R1A5A6000840, RS-2026-25488333). Financial support from the Institute for Research in Finance and
Economics of Seoul National University is gratefully acknowledged.
References
- Akhtari and Li, (2024) Akhtari, B. and Li, H. (2024). The Cox–Ingersoll–Ross process under volatility uncertainty. Journal of Mathematical Analysis and Applications, 531(1):127867.
- Bernanke and Woodford, (1997) Bernanke, B. S. and Woodford, M. (1997). Inflation forecasts and monetary policy. Journal of Money, Credit and Banking, 29(4):653–684.
- Brunnermeier and Sannikov, (2014) Brunnermeier, M. K. and Sannikov, Y. (2014). A macroeconomic model with a financial sector. American Economic Review, 104(2):379–421.
- Denis et al., (2011) Denis, L., Hu, M., and Peng, S. (2011). Function spaces and capacity related to a sublinear expectation: application to G-Brownian motion paths. Potential Analysis, 34:139–161.
- Epstein and Ji, (2013) Epstein, L. G. and Ji, S. (2013). Ambiguous volatility and asset pricing in continuous time. The Review of Financial Studies, 26(7):1740–1786.
- Fleming and Soner, (2006) Fleming, W. H. and Soner, H. M. (2006). Controlled Markov processes and viscosity solutions. Springer.
- Gao, (2009) Gao, F. (2009). Pathwise properties and homeomorphic flows for stochastic differential equations driven by G-Brownian motion. Stochastic Processes and their Applications, 119(10):3356–3382.
- Hölzermann, (2021) Hölzermann, J. (2021). The Hull–White model under volatility uncertainty. Quantitative Finance, 21(11):1921–1933.
- Hölzermann, (2022) Hölzermann, J. (2022). Term structure modeling under volatility uncertainty. Mathematics and Financial Economics, 16(2):317–343.
- Hölzermann, (2024) Hölzermann, J. (2024). Pricing interest rate derivatives under volatility uncertainty. Annals of Operations Research, 336(1–2):153–182.
- (11) Hu, M., Ji, S., Peng, S., and Song, Y. (2014a). Backward stochastic differential equations driven by G-Brownian motion. Stochastic Processes and their Applications, 124(1):759–784.
- (12) Hu, M., Ji, S., Peng, S., and Song, Y. (2014b). Comparison theorem, Feynman–Kac formula and Girsanov transformation for BSDEs driven by G-Brownian motion. Stochastic Processes and their Applications, 124(2):1170–1195.
- Hu and Peng, (2021) Hu, M. and Peng, S. (2021). Extended conditional G-expectations and related stopping times. Probability, Uncertainty and Quantitative Risk, 6(4):369–390.
- Hu and Wang, (2018) Hu, M. and Wang, F. (2018). Ergodic BSDEs driven by G-Brownian motion and applications. Stochastics and Dynamics, 18(6):1850050.
- Hu and Peng, (2009) Hu, M.-s. and Peng, S.-g. (2009). On representation theorem of G-expectations and paths of g-Brownian motion. Acta Mathematicae Applicatae Sinica, English Series, 25(3):539–546.
- Hu et al., (2018) Hu, Y., Lin, Y., and Soumana Hima, A. (2018). Quadratic backward stochastic differential equations driven by G-Brownian motion: Discrete solutions and approximation. Stochastic Processes and their Applications, 128(11):3724–3750.
- Hu et al., (2022) Hu, Y., Tang, S., and Wang, F. (2022). Quadratic G-BSDEs with convex generators and unbounded terminal conditions. Stochastic Processes and their Applications, 153:363–390.
- Hu and Zhou, (2019) Hu, Z. and Zhou, Q. (2019). Convergences of random variables under sublinear expectations. Chinese Annals of Mathematics, Series B, 40(1):39–54.
- Kim and Park, (2024) Kim, J. and Park, H. (2024). A G-bsde approach to the long-term decomposition of robust pricing kernels. arXiv preprint arXiv:2409.00535.
- Kiyotaki and Moore, (1997) Kiyotaki, N. and Moore, J. (1997). Credit cycles. Journal of Political Economy, 105(2):211–248.
- Li and Peng, (2011) Li, X. and Peng, S. (2011). Stopping times and related itô’s calculus with G-Brownian motion. Stochastic Processes and their Applications, 121(7):1492–1508.
- Lin et al., (2026) Lin, Y., Sun, Y., and Wang, F. (2026). Infinite horizon quadratic backward stochastic differential equations driven by G-Brownian motion. Electronic Journal of Probability, 31:1–37. Paper No. 112.
- Peng, (2019) Peng, S. (2019). Nonlinear expectations and stochastic calculus under uncertainty: with robust CLT and G-Brownian motion, volume 95. Springer Nature.
- Sun and Wang, (2024) Sun, Y. and Wang, F. (2024). G-forward performance process and representation of homothetic case via ergodic quadratic G-BSDE. Probability, Uncertainty and Quantitative Risk, 9(1):85–106.
- Vorbrink, (2014) Vorbrink, J. (2014). Financial markets with volatility uncertainty. Journal of Mathematical Economics, 53:64–78.
- Xu et al., (2011) Xu, J., Shang, H., and Zhang, B. (2011). A Girsanov type theorem under G-framework. Stochastic Analysis and Applications, 29(3):386–406.