Search
K

Appendix: Math

Setup

Each day we have five relevant values:
  • Yesterday’s asset prices in dollars, p’
  • Today’s asset prices in dollars, p
  • Yesterday’s (human) quantity vector, q’ on a per LP token basis
  • Today’s (human) quantity vector, q on a per LP token basis
  • Our target wealths vector, v, which we take to be wealth fractions:
ivi=1   vi0\sum_i v_i = 1 \ \ \ v_i \geq 0
Observe that v should tend to be proportional to the actual wealth (q_i*p_i) on each asset i.

Calculating Daily Returns & Benchmarks

All of these benchmarks are designed so that 0 represents no change. Since the values are expected to be small, they may not need to have log taken for additive basis points.
Daily $ Returns
ipiqiipiqiipiqi\frac{\sum_i p_iq_i - \sum_i p'_iq'_i}{\sum_i p'_iq'_i}
This measures how much the portfolio increased in $ terms since yesterday.

Daily Crypto Basis Returns

ipiqiipiqiipiqi\frac{\sum_i p_iq_i - \sum_i p_iq'_i}{\sum_i p_iq'_i}
This value will be negative if we would prefer yesterday’s portfolio to today’s portfolio at today’s prices.

Rebalanced Portfolio Benchmark

i(pipi)vi1\sum_i \left(\frac{p_i}{p'_i}\right) v_i - 1
This is our target benchmark for Clipper: our daily dollar gain should track it closely and ideally outperform.
Observe that a rebalanced portfolio can outperform a static (”held”) portfolio consisting of any (!) initial combination of the assets themselves. For instance, imagine the two-asset case consisting of a risky asset and a stablecoin. On day one, the risky asset doubles in price. On day two the risky asset returns to its original value. It can be verified that a 50-50 rebalanced portfolio will return 1.125x, vs. 1x for simply holding any combination of assets for the two days.

CPMM Loss Benchmark

In the CPMM, we have that:
qivi=Civilogqi=logC\begin{align*} \prod q_i^{v_i} & = C \\ \sum_i v_i \log q_i & = \log C \end{align*}
Day-to-day, the portfolio of the CPMM changes to have wealth proportional to the new set of prices. By working out this constant log-sum, we get the following expression for how much “impermanent” loss a CPMM mechanism takes on:
i(pipi)vii(pipi)vii(pipi)vi\frac{\prod_i \left( \frac{p_i}{p'_i} \right)^{v_i} - \sum_i \left( \frac{p_i}{p'_i} \right)v_i}{\sum_i \left( \frac{p_i}{p'_i} \right)v_i}
This value is at most 0 when the prices are the same. Clipper’s crypto performance (loss or gain) should substantially outperform this benchmark.