Title: The Liquidity Risk Pricing
1The Liquidity Risk Pricing
- Andrey Marchenko
- GGY Inc., (AndreyMarchenko_at_GGY.com)
2Liquidity definition
- A seller wants to sell a volume V0 of common
stock (in stock units)
3Liquidity model
- The more asset units he attempts to sell at the
moment, the lower will be the price per unit,
therefore ?p/?a lt 0 (for sales). - Liquidity measure can be defined as l -?p(t,
a)/?a the greater is l, the lower is the
liquidity. - If l 0 the price does not depend on the asset
flow no friction market with absolute liquidity
case.
4Some experimental data
5The problem
- Selling strategy the flow of sells u(t)
generates cash flow p(t,
a(t)) a(t)dt - Its present value equals
- (1)
- Total volume sold equals
- (2)
- Find
- Maxa sa over strategies a(t) or its
expectation for random p(t, a(t)) - The optimal strategy a(t)
- under restriction (2)
- Liquidity price V0 p(0, 0) - Maxa sa
6Different problems
- Deterministic problem
- non-random stock price p(t, a) known in advance.
- Stochastic problem p(t, a) ? p(? t, a)
- Static strategy a prefixed selling strategy in
stochastic environment. It can be used by
regulators in order to evaluate and limit
possible losses. - Dynamic strategy - optimal control approach
assumes dynamic strategy depending on a current
state of the random stock price.
7Deterministic problem analysis
- If
- p(t, a) p(a),
- d 0,
- Borrow V0 p(0) dollars at t 0 and sell the
asset infinitely slow at a price p(0) max
ap(a). - So the liquidity price 0.
- If -?p(t, a)/? a 0 (absolute liquidity) p (t,
a(t)) p(t) the optimal strategy is selling all
the stock V0 at a moment t0 when p(t0) max t
p(t) - In both cases gain V0 max p.
- Therefore assume l, d gt 0.
8Generic solution formalism
- Standard variation calculus yields
- with is the (unknown) Lagrange coefficient z .
- Solving with respect to a we get a family in z of
functions a(z , t). - Condition
- adds an equation. Eliminate z and substitute to
(3) to get s s (T). - Find max Ts by simply differentiating.
9High liquidity approximation
- Assume for simplicity
- p(0, 0) 1 V0 1 d 1
- High liquidity p(t, a) p(t) - La/2
- Stationary liquidity L const
10Generic solution
- Assume L const then
- a(t) (p(t) - zet) / L.
- L z (eT - 1),
- Solve for ds/dT 0 in T and substitute to
obtain s(L)
11Example linear asset price
- Assume p(t) 1bt . Then
- and
12Graphs of s(L, b)
13Comparison with simplest optimization
14Stochastic problem
- p(t, a) p(? t, a), ? being random parameter.
- Three approaches 3 problems
- Static strategy
- The seller is risk-neutral
- The seller is risk-sensitive
- Dynamic strategy
15Stochastic problem static a risk-neutral
seller
- Interested in maximization of Esa
- Since a(t) is not random, the Fubini theorem
yields - and reduces the problem to the deterministic one
with Ep(t, a) substituting p(t, a)
16Stochastic problem static a risk-sensitive
seller
- Choose the risk metric and plot all admissible
points in the Risk-Return plane - Find the optimal point on the effective boundary
maximizing utility function - Analogy to the classic Markowitz theory
- portfolio choice ? sales allocation to different
times a(t) - return Esa
- risk metric standard deviation of sa
17Static a risk-sensitive seller 2
- Assume high liquidity approximation
- p(? t, a) p(? t) - La/2 with m (t)
Ep(? t) - return x
-
- risk
- measure y
- where C(t1, t2) Ep(t1) p(t2) - E p(t1)
Ep(t2)
18Static a risk-sensitive seller 3
- Effective boundary equation
- dReturn/da 0.5kdRisk measure/da z
19Static a risk-sensitive seller 4
- These equations have unique solution for fixed T
gt 0 and - min0, T C(t, t)e-2t lt D lt max0, T C(t,
t)e-2t - The practical way to is to apply Galiorkin
approximations and get a finite dimensional
problem.
20Example
- Let dp p(ndt s dW) in risk-neutral world
- m(t) expnt,
- C(t1, t2) exps2(t1 t2)/2( exps2min(t1,
t2) 1) - Galiorkin approximations
- Order 0 a(t) ? 0, T (t) /T
- Order 1 a(t) ? 0, T (t)a0 a1 t /(a0 T
a1 T 2/ 2)
21Example - results
22Stochastic problem - Dynamic strategy optimal
control
- Price process dp p(ndt s dW)
- Control dx -a(t) dt
- t time x (t) unsold stock a(t) ? 0.
- Optimized functional
- Boundary conditions
- p(0) 1 x(0) 1.
23Hamilton-Jacobi-Bellman equation
- Assume high liquidity approximation
- p(? t, a) p(? t) - La/2
- Hamilton-Jacobi-Bellman equation for V
- With t, ? ? 0 and 0 ? z ? 1
- max a- ?V/? z a (? - La /2)a e-t (et ?V/? z
- ?)2e-t /2L - Boundary conditions are V 0 for z 0 and for
? 0
24Equation for V
- Final ultra-parabolic equation
- For stationary process (µ and s do not depend on
t) a substitution V(?, ?, t) U(?, ?) ? e-t
leads to nonlinear parabolic equation
25Some problems to investigate
- Change price process
- Add random risk free rate
- Modify the liquidity model
- Change p(t, u(t))
- Consider random liquidity
- Complex liquidity models, modeling trade patterns
and behavior - Investigate time-discrete sales