- CLS 1 hour aggregated price feed (Jan 2018 onwards)
- CLS 1-hour volume price feed
- CLS Flow feed (taking out inter-dealer trades)
- MUFG aggregated order book data
- We filter out any CLS inter-dealer flow data and keep the non inter-dealer flows. This is because non-dealers like corporates, hedge funds and non-bank financial institutions have limited or no access to the wholesale markets when compared to dealers.
- Liquidity is modelled using machine learning as a multi-factor time series using various factors including market impact, normalised buy-sell imbalance, percentage of gross volume settled in hourly buckets and autocorrelation.
- For each state, the above liquidity factors were constructed from CLS data.
- The cost of execution was derived from the aggregated order book data with a sweep size typical of a given currency pair.
- We then fit a non-linear model to the cost of execution for each market/currency pair.
- Liquidity scores across different states are then normalised between 0 and 1, where higher scores are better.
What are the COVID & Pre-COVID periods chosen for analysis?
- PRE-COVID: 1st June 2019 – 20th February 2020
- RECENT: 13th October – 3rd November
- COVID PEAK: 21st February – 20th March
How do we define market impact?
- Market impact is defined as the effect that a buying or a selling activity has on the volatility-adjusted returns in hourly buckets.
What is volume imbalance?
- Volume imbalance is defined as the imbalance between supply and demand at the best bid and ask prices.
Can we predict liquidity term structure?
- Yes. The initial analysis is purely contemporaneous explanation. Phase 2 of this project will be forecasting 24 hours look-ahead liquidity term structure evolution.
Can we add impact of major economic events on the liquidity term structure?
- Yes. We are looking at extending the model to capture influence of key economic events to liquidity.