The TRADE 63 - Q1 2020 | Page 13

[ T H O U G H T L E A D E R S H I P | G O L D M A N S A C H S ] Figure 2: Stocks A and B broken by traditional and clusters view, left and right, respectively. Stock B: VWAP, OTD, Max 5% Stock A : VWAP, OTD, Max 5% Traditional Market View • FTSE 100 • Large Cap • Index Name Cluster 1 • The most liquid of all • Consistent and predictable Traditional Market View • FTSE 100 • Large Cap • Index Name Cluster 3 • Price and volume spikes • Information leakage Source: GS Global Markets Division. For Illustrative purposes only. Figure 3: Shifts in intraday correlation and volatility for S&P500 stocks. 70% 50% United States - Correlation United States - Volatility 45% 60% 40% 50% 35% Securities Division | Confidential 40% 30% 2 30% 25% 20% 15% 20% 10% 10% 5% 0% Jan 2014 09:30 - 10:00 Jan 2015 10:00 - 11:00 Jan 2016 11:00 - 12:00 Jan 2017 12:00 - 15:00 Jan 2018 15:00 - 16:00+ 0% Jan 2014 09:30 - 10:00 Jan 2015 Jan 2016 10:00 - 11:00 11:00 - 12:00 Jan 2017 12:00 - 15:00 Jan 2018 15:00 - 16:00+ Source: Goldman Sachs QES, Reuters, Bloomberg the simple ‘addition’ of single stock variances. Over the last couple of years our team has incorporated many of the above insights, and a plethora of oth- ers, to our suite of Algorithmic Port- folio Execution products (APEX) to Securities Division | Confidential help our clients improve execution and reduce slippage. When we consider execution of anything more than a single stock, we utilise three modelling pillars to account for intraday (and overnight, where applicable) dynamics: risk, impact and multi-period optimisation. Risk modelling accounts for the non-stationary nature of the cova- riance structure of stocks intraday and for overnight effects explicitly. Market impact estimation is more granular—it incorporates time of day effects and models are trained within a more homogenous set of stock characteristics to improve their explanatory power. Finally, optimisation of the balance between risk and cost is multi-period, dynam- ic in its horizon specification, and adaptive to the characteristics of the sets of stocks being traded and the execution objective. Portfolio execution is a complex area of trading that has historically been approached with similar theo- rems and practices utilised in quant portfolio management, like factor models and mean-variance opti- misation tools, typically ingesting daily data The ever-changing nature of global markets, the non-linear convolution of micro and macro effects, the non-stationary nature of the higher frequency covariance structure, the global intraday con- tagion effects and the continuously morphing impact of exogenous factors—like the active to passive shift—onto trading dynamics, render such portfolio approaches, as well 1 as purely single-stock calibrated algorithms, relatively ineffective in trading portfolios of stocks. The preponderance of computing power, coupled with innovative modelling techniques and excru- ciating curation and interrogation of masses of data, have allowed us to take a few steps closer —we believe—to deciphering some of the more pertinent questions of optimal execution. The beauty (and bane) of research in our industry is that this quest is fairly boundless in its scope. The above information is an analysis under certain scenarios and is not a representation of future performance. The stated values above are the theoretical as of the date calculated and are not representation that any transaction can be effected at that values. We are not soliciting any action based upon this material. The information provided herein was supplied in good faith based on information which we believe, but do not guarantee, to be accurate or complete; however we are not responsible for errors or omissions that may occur. The information provided is historical data. Past performance is not indicative of future results. For illustration purposes only. Issue Issue 63 63 // // thetradenews.com thetradenews.com // // 13 13