Mahi Methodology
Mirroring and tweaking scientific method, the Mahi methodology is to first:
Capture information relating to a measured behaviour
without capturing the information in the first place, you do not know how much of an impact it will be having on your business
Visualise information relating to the measured behaviour, and at different levels of granularity:
on a per tick basis e.g. Echo Top Of Book or Short Simulations
on a per trade basis e.g. Echo Yield Profiles
on a rolling minute basis e.g. Trading Dashboard Analytics
in aggregate over the trading week e.g. Pricing Predictivity
over months/years of analysis e.g. Long Simulations and Account Summary
Manual or automated analysis of the information
do patterns emerge?
can machine learning make sense of the problem?
how can this be optimised?
are there any proxies for a particular outcome? predictivity => pnl. flow classification => pnl
Hypothesis formation
A/B or multivariate testing
beta models
it's requires way less capital to simulate a problem (if this is feasible) relative to conducting the same learnings in production
Proactively monitor responses relating to the measured behaviour and alert users if there is potential harm.
offering automated views to assist the decision making and make action more swift
one click actions
Automate management of responses to the measured behaviour.
agents reporting on their overall performance
self-optimising
rehabilitating incorrect decisions
support for manual overrides (but largely unnecessary)
The above is a process whereby each stage in the process can trigger downstream refinements of our methodology and applied automation techniques to the category of analysis.
Software & Analysis
We make changing our system easy. This reduces the time to action a code change response to observed behaviours. To make this safe we use:
automated testing
simulation backtesting (all production code is reused between live and backtest)
can update the system without client impact multiple times a day, safely.