Capturing edge in the race to efficient execution
Transaction Cost Analysis (TCA) plays an important role in defining execution strategy in all environments within the trading arena.
Efficient trade execution can save clients and investors millions of dollars a year, dramatically improving the performance of fixed income portfolios and actively managed funds.
Evolving technology in a challenging marketplace
At S&P Global Market Intelligence, a leading provider of fixed income market data solutions, we are placed at the forefront to offer artificial intelligence (AI) technology, providing essential insight and analysis for single entities through to full portfolio pre-execution analysis, irrespective of the market environment.
The onset of state-of-the-art cloud-based technology, contributor-based in-house data and cutting-edge machine learning technologies, provides our subscribers with essential analytics tools to stay ahead of the curve.
Dynamic modelling for a changing landscape
Our AI utilises a learning process which identifies changing market dynamics to select the best model to estimate execution pricing and slippage, providing predictions and criteria for optimal execution with respect to single traded entities through to complex baskets, based on real-time market dynamics:
Case study
A data sample from 2020
We are using a sample of 88,460 trades from 1,669 unique investment grade ISINs split between 255 US treasuries, 167 global sovereign bonds and 1,247 global corporate bonds from S&P Global Ratings B to AAA, inclusive. There is a relatively even weighting of 42,805 buys and 45,655 sells in the case study period:
Overview
The aim of S&P Global Market Intelligence Fixed Income Pre-trade is to provide essential and valuable tools to improve transaction cost estimation and execution potential by means of understanding the dynamics behind efficient execution in changing markets.
Whilst the goal in OTC bond trading is to achieve buys and sells as near to bid and offer prices respectively, the mid-price quote will be used as the neutral marker. Costs in the form of price slippage will be represented as a negative value and conversely, any improvement in mid-price will have a positive value.
Assumptions:
- Mid-price is calculated as the arithmetic average between best bid / offer from the average of RFQs supplied by our Price Viewer
- Slippage is defined as the difference between executed price and price viewer mid-price Credit ratings are provided by S&P Global Bond Ratings\
- Total nominal traded size equal to $530 billion
- All data is converted into a base currency of US dollars for comparability purposes
Density function of market execution slippage
Using currency-adjusted execution price as the standardiser of the slippage value from our Price Viewer mid-price, 39.13% were executed better than mid-price, comprising 41.78% of buy orders versus a slightly lower 36.78% of sell orders.
The negative skew is very apparent, driven by this poorer performance in the period by sell orders.
Overall, trades have been executed poorly based on an average slippage of 0.17% of execution price at the cost of $0.15 versus mid-price.
Zooming in on the interquartile ranges with the Box Plot illustration shows the tighter execution between the 1st and 3rd quartile ranges, with a clear tighter range between absolute investment grades, UST, A and B.
Noise within the tiers is expected depending on the dataset, Covid-impacted appetite for risk in 2020 and real-time liquidity.
Outliers in the dataset confirm that better execution could have been established with huge cost cutting results.
Fixed income pre-trade
An adaptive model for prediction
Changing market characteristics and drivers warrant adaption as essential in any predictive model, whether estimation is required from a relative or independent perspective.
Removing constraints and supervision unleashing the full capabilities of AI to self-calibrate without the need for endless parameter optimisation and targeting.
S&P Global Market Intelligence Fixed Income Pre-trade has a single goal throughout its selection process, to achieve a model that minimises predictive error in real-time.
Any number of inputs (independent variables) can be used in the calculation process, the significance determined internally, and the optimal predictor achieved.
Exploring the percentage of variance explained
By transforming available market data into a standardised space, eliminating any scale bias we can determine the significance of inputs and their contribution to defining regimes, thus removing any issues of multi-collinearity along with isolating market real-time drivers.
**The above percentage of variance illustrations are taken from random ISINs within the selected ratings and maturities from the 2020 dataset.
The transformed coordinates not only resonate in components with consistently similar characteristics per rating tier but also highlight trending drivers within maturities.
Maximising percentage of variance explained at the same time as minimising prediction error is the goal of achieving the best estimator.
Mean-squared error and r-squared: the quality controllers
To value the improvement in prediction by using an adaptive learning model, we can consider the modelling accuracy by means of the normalised by variance statistic, R-Squared.
Below shows the relevant R-Squared values between variable dimensions and that of a selective mean- squared error minimising model (labelled as ‘R-Squared_Mininised’) within our machine learning process:
The benefit of an adaptive error-minimising model versus a predefined linear input model is clear: the result being the adaptive model efficiently valuing predictions for execution price and therefore slippage costs in real time.
Estimating size impact of execution
Using beta coefficients for a single ISIN, calculating the buy and sell aggressor estimations for an A- rated, high grade corporate bond, we can see the changing impact on the execution price and the cost estimation due to changes in execution quantity alone, by using static input values for all coefficients except the execution quantity.
For this bond, the relationship within the dataset produces a near linear relationship between quantity traded and execution price, when all other input variables are held the same.
For the smaller end of the quantity scale, it indicates there is a possibility of small gain in buy execution versus mid, ie. execution closer to bid-price, which would be regarded as executing at mid-price from a risk averse aspect, but slippage increases at a gradient as buy quantity increases.
In terms of sell execution, the situation is quite contrary, the impact of traded quantity being absorbed, with negligible impact to execution price as quantity increases. This market could be described as well- bid.
The benefits of an adaptive learning model not only end at TCA prediction, but the generic build of the model enables the end-user to select inputs and outputs for any complex baskets to achieve a variety of essential pre-trade analysis, including a representation of current market state.
Identifying inherent dimensionality characteristics will undoubtedly improve returns and provide insight in changing market dynamics ahead of the curve.
Contact us for more information
S&P Global Market Intelligence
At S&P Global Market Intelligence, we understand the importance of accurate, deep and insightful information. Our team of experts delivers unrivaled insights and leading data and technology solutions, partnering with customers to expand their perspective, operate with confidence, and make decisions with conviction.
S&P Global Market Intelligence is a division of S&P Global (NYSE: SPGI). S&P Global is the world’s foremost provider of credit ratings, benchmarks, analytics and workflow solutions in the global capital, commodity and automotive markets. With every one of our offerings, we help many of the world’s leading organisations navigate the economic landscape so they can plan for tomorrow, today. For more information, visit www.spglobal.com/marketintelligence.
Copyright © 2022 by S&P Global Market Intelligence, a division of S&P Global Inc. All rights reserved. No content, including by framing or similar means, may be reproduced or distributed without the prior written permission of S&P Global Market Intelligence or its affiliates. The content is provided on an “as is” basis.