Drawdown and Its Linearity

Optimization
Risk
Conditional drawdown at risk, its special cases, and how to make the optimization linear using an auxiliary variable.
Author

Manu Rathi

Published

December 1, 2022

In this article, we first discuss the conditional drawdown at risk (CDaR), its special cases — maximum drawdown and average drawdown — and then move to the optimization problem for the return–CDaR efficient frontier as discussed by Uryasev et al. 1

Assumption

This approach concentrates on the portfolio equity curve over a particular sample path (historical, or the most probable future sample path) — i.e. we define a sample-path risk function rather than a risk measure over a set of sample paths. It makes no assumption about the underlying probability distribution, which unlocks various practical applications.

CDaR

For some value of the confidence parameter \alpha, the \alpha-CDaR is defined as the mean of the worst (1-\alpha) \times 100\% drawdowns experienced over some period of time on a sample path.

The CDaR risk function is based on the CVaR concept and can be viewed as a modification of CVaR to the case where the loss function is defined as a drawdown. Similarly to CVaR — the average of all losses greater than or equal to \alpha-VaR — CDaR is the average of all drawdowns greater than or equal to the \alpha drawdown-at-risk (\zeta_\alpha), such that a low CVaR ensures a low VaR.

Risk functions are parameterized by the portfolio weight matrix W; T is the total time period considered for the calculation — the sample-path time horizon.

CDaR_\alpha(W) = \frac{1}{(1- \alpha)T} \int_{\Omega_\alpha} D(W,t) \,dt \tag{1}

where \Omega(W) = \{t \in [0,T] : D(W,t) \geq \zeta_\alpha(W)\}.

By definition, with a specified probability \alpha, the \alpha-DaR (\zeta_\alpha) of a portfolio is the lowest amount such that the probability the drawdown will not exceed \zeta_\alpha is greater than or equal to \alpha.

We consider two risk functions that are special cases of CDaR — maximum drawdown (MaxDD) and average drawdown (AvDD) — which are limiting cases as \alpha \to 1 and \alpha \to 0 respectively:

MaxDD(W) = \max_{t \in [0,T]} D(W,t) \tag{2}

AvDD(W) = \frac{1}{T} \int_{0}^{T} D(W,t) \,dt \tag{3}

Linear optimization

MaxDD

We consider a modification of the classic return–risk efficient-frontier optimization: maximize return while constraining maximum drawdown to a pre-defined level.

\max_{W}{W^TR}

constrained on

MaxDD(W) \leq v_1 C

where R is the return matrix R = [r_1, r_2, r_3, \dots, r_\gamma], \gamma is all constituents, and v_1 is the percentage of deployed capital C allowed to be in drawdown.

Maximum drawdown is not linear in the parameter W. It can be made linear using an auxiliary variable, as follows:

MaxDD(W) : \max_{k \in (0,T)} DD_k \leq v_1 C \tag{4}

MaxDD(W) : \max_{k \in (0,T)} \Big( \big(\max_{j \in (0,k)} W^TR_j\big) - W^TR_k \Big) \leq v_1 C \tag{5}

Converting the first maximum function in Equation 5, using the property of maximum, applies a constraint on each day to be less than the desired drawdown:

MaxDD(W): \big(\max_{j \in (0,k)} W^TR_j\big) - W^TR_k \leq v_1 C \quad \forall\ k \in [0,T] \tag{6}

Converting the second max function:

MaxDD(W) : u_k - W^TR_k \leq v_1 C \quad \forall\ k \in [0,T] \tag{7}

where u_k = \max_{j \in (0,k)} (W^TR_j) is an auxiliary variable.

Defining u_k in Equation 7: u_k is equal to a maximum function, hence it’s an increasing function.

u_k \geq u_{k-1} \quad \forall\ k \in [0,T],\ \text{with } u_0 = 0 \tag{8}

Also, u_k should be equal to or greater than W^TR_j:

u_k \geq W^TR_j \quad \forall\ j \in [0,k],\ k \in [0,T] \tag{9}

W_{min} \leq W \leq W_{max} \tag{10}

So the optimization is converted to a convex optimization problem with a linear performance function and a piecewise-linear convex constraint, given by maximizing return subject to Equation 6 through Equation 10.

CDaR

\max_{W}{W^TR}

constrained on

CDaR_\alpha(W) \leq v_2 C

Rewriting Equation 1:

CDaR_\alpha(W) = \min_{\zeta}\Big[\zeta+\frac{\sum_{k \in (0,T)}[DD_k-\zeta]^+}{(1-\alpha)T}\Big] \tag{11}

where the function [x]^+ is defined as \max(0,x).

The CDaR constraint function can be linearized as follows:

\zeta+\frac{\sum_{k \in (0,T)}z_k}{(1-\alpha)T} \tag{12}

Defining z_k:

z_k = [DD_k-\zeta]^+

Since z_k is the positive part:

z_k \geq DD_k - \zeta \quad \forall\ k \in [0,T] \tag{13}

Using Equation 7, DD_k can be written in terms of u_k:

z_k \geq u_k - W^TR_k - \zeta \quad \forall\ k \in [0,T] \tag{14}

And since z_k is always positive:

z_k \geq 0 \tag{15}

Similarly to MaxDD, the optimization problem converts to maximizing return subject to Equation 14, Equation 15, and Equation 8 through Equation 10.

An important feature of this formulation is that it does not involve the threshold function \zeta_\alpha directly. At the optimal solution, the variables W and \zeta_\alpha together give the optimal portfolio and the corresponding value of the threshold function — see 2 for further discussion.

Footnotes

  1. Alexei Chekhlov, Stanislav Uryasev, Michael Zabarankin. Portfolio Optimization With Drawdown Constraints. ise.ufl.edu/uryasev↩︎

  2. R. Tyrrell Rockafellar, Stanislav Uryasev. Optimization of Conditional Value-at-Risk. ise.ufl.edu/uryasev↩︎