Objectives

We include a few simple utility functions with this package. Arbitrary utility functions (which can include constraints) are specified in the objectives.jl file. We outline the included utility functions and how to specify new ones below.

Included utility functions

LinearNonnegative (arbitrage)

The LinearNonnegative objective is

\[ U(\Psi) = c^T\Psi - \mathbf{I}(\Psi \geq 0),\]

where $c$ is a positive price vector. $\mathbb{I}(\Psi \ge 0)$ is an indicator function that is $0$ if $\Psi \ge 0$ and $+\infty$ otherwise. This objective requires no net input and a linear utility for each token, defined by the price vector. A nonzero solution to this problem finds an arbitrage in the network—a set of trades that yields strictly positive value, but requires no net input.

BasketLiquidation (liquidations and swaps)

The BasketLiquidation objective is

\[ U(\Psi) = \Psi_i - \mathbf{I}(\Psi_{-i} + Δ^\mathrm{in}_{-i} = 0, ~ \Psi_i \geq 0),\]

where $Δ^\mathrm{in}$ is a basket of tokens to be liquidated into token $i$. Here, $\Psi_{-i}$ is the vector $\Psi$ with entry $i$ removed, and $\mathbf{I}(\Psi_{-i} + Δ^\mathrm{in}_{-i} = 0, ~\Psi_i \geq 0)$ is the indicator function which is zero if all conditions are met and is $\infty$, otherwise. In the special case where $Δ^\mathrm{in}_k$ is zero at all indices except for some index $j \ne i$, this objective defines a swap from token $j$ to token $i$ where we attempt to maximize the amount of token $i$ received. We implement a Swap objective as shorthand for this special case.

Specifying new utility functions

New utility functions can be easily specified following the examples in objectives.jl. We only need to specify the conjugate $f(\nu)$ and its gradient:

\[ f(\nu) = \sup_\Psi \left(U(\Psi) - \nu^T \Psi \right)\]

and

\[\nabla f(\nu) = -\Psi^\star,\]

where $\Psi^\star$ is the optimal value of the supremum in $f(\nu)$. In addition, we specify lower bound lower_limit and upper bound upper_limit for the objective.

Sometimes $f(\nu)$ does not have a closed form solution, so we need to solve an additional optimization problem to evaluate the objective. For example, consider Markowitz portfolio rebalancing:

\[U(\Psi) = \mu^T(\Psi + \Delta^\mathrm{in}) - \frac{\gamma}{2} (\Psi + \Delta^\mathrm{in})^T\Sigma(\Psi + \Delta^\mathrm{in}) - \mathbb{I}(\Psi + \Delta^\mathrm{in} \geq 0),\]

where $\mu$ and $\Sigma$ are the (estimated) mean and covariance returns for each token.