ADAPTIVE Intelligence

CLSA’s ADAPTIVE, a revolution in Algorithmic Trading integrates the latest AI machine learning technology with contextual news and research. The supercharged CLSA Algorithmic Trading platform now enables our clients to take a more discreet, proactive and adaptive approach to execution. As non-traditional market participants continue to exert their influence on the Asian trading landscape, CLSA’s ADAPTIVE levels the playing field for our clients in an ever-changing and increasingly volatile trading environment.

CLSA’s ADAPTIVE leverages the Neural Network, the machine learning technology behind ‘The Brain’ which utilises both public and proprietary data sources to train itself to identify volatility and predict intra-day price movements as well as market volume. Real-time trend momentum evaluation is supported by confidence levels and intensity screening of predicted price trend and market volume.

Live news feed analysis powered by a Natural Language Processing engine allows ADAPTIVE to detect and react to momentum shifts as they occur. Using diverse data sources including Bloomberg, Google Finance, Twitter and CLSA’s award-winning research, ‘The Brain’ classifies data into five different sentiment levels applying contextual interpretations to live markets.

Anatomy of ADAPTIVE

Price & volatility signals/volume

prediction

Sentimentsignals

ADAPTIVE Intelligence

A Revolution in Algorithmic TradingADAPTIVE

Time to ADAPT

Increasingly volatile markets require accurate, effective and timely analysis of both data and momentum.

CLSA’s ADAPTIVE addresses these requirements in order to ensure our clients remain on the front foot as markets move, whilst easing the impact from disruptive non-traditional market participants. By differentiating between long-term sentiment changes and short-term momentum moves, CLSA algorithms take advantage of momentum moves but also of mean reversion. Combinations of predictions are generated to boost the performance of strategies targeting various benchmarks.

ADAPTIVE generates price trend signals and performs real time volatility clustering (measurement) for its decision making. ADAPTIVE predicts short-term price, volume and projected target (i.e. VWAP projection) to adjust strategies’ behaviour. Every signal is classified by a confidence level and intensity grade based on real-time trades. At the same time, ADAPTIVE makes long-term forecasts (i.e. closing volume prediction) and engineers the ‘market direction indicator’ to detect the presence of informed trading in the market. With directional market information, ADAPTIVE algorithms can dynamically alter the execution plan based on favourable or adverse market moves. The trade execution plan is dynamically revised throughout the trading day to adapt to market conditions as they evolve.

Implications for algorithmic trading strategies

A Revolution in Algorithmic TradingADAPTIVE

For more information contact clsaee@clsa.com clsa.com/liquidity

DISCLAIMER: This publication is for information purposes only and is not a solicitation or any offer to buy or sell. It is not intended to provide professional, investment or any other type of advice or recommendation. We expressly disclaim all warranties, express or implied of any kind relating to this publication including but not limited to the warranties of merchantability, fitness for a particular purpose and the performance of algorithmic trading,which may be subject to system failure, telecommunication error, etc. You should discuss how particular trade(s) affect you, with your professional tax, legal, or other adviser. This publication is distributed into the United States solely to persons who qualify as “Major US Institutional Investors” as defined in Rule 15a-6 under the Securities and Exchange Act of 1934, as amended, and who deal with CLSA Americas, LLC.Different jurisdictions may impose particular regulatory restriction over algorithmic trading, which shall prevail over this publication. The content of this publication is subject to CLSA’s Legal and Regulatory Notices as set out at www.clsa.com/disclaimer.html.

May 2018

Market % of time active trend signal detected

Up trend % Down trend % Signal prediction accuracy %

Actual market up trend price move (bps)

Actual market down trend price move (bps)

Japan 39.23 28.06 11.16 89.31 3.53 - 10.07 3.41 - 10.40

India 30.73 21.07 9.66 86.01 2.26 - 7.32 2.40 - 8.51

Hong Kong/Australia

40.05 28.28 11.77 96.26 1.90 - 4.60 2.10 - 4.70

ADAPTIVE PRICE PREDICTION STATISTICS FOR MAJOR MARKETS

Note: Statistics compiled from all orders traded with CLSA ADAPTIVE algorithms in 1Q 2017.

ADAPTIVE VWAP (VWAP)ADAPTIVE STRATEGIES

Time

SIGNALS

71.7

10:00

Price

11:00 12:00 13:00

71.9

71.8

72.0

72.2

72.1

72.3

The analytical engine generates trend signals characterised by intensity based on real time trades. It also constructs a volatility band (the blue lines) around the stock price, highlighting normal vs. abnormal (favourable or unfavourable swings) volatility.

Traditional VWAP execution profile (blue line) vs. ADAPTIVE VWAP (red line) trading around the VWAP curve based on ADAPTIVE signals.

Time

EXECUTION PROGRESS%

Traditional VWAP execution profile

ADAPTIVE VWAP

With ADAPTIVE, strategies such as VWAP no longer need to stick to a traditional time-based fixed slicing schedule. It now reacts to real-time market price and volume movements. Price and volatility signals enable the VWAP strategy to capture small price misplacements in the market. With a dynamic view of future forecast volumes and price levels, ADAPTIVE VWAP achieves outperformance with reduced risk.

DISCLAIMER: This publication is subject to and incorporates the terms and conditions of use set out on www.clsa.com/disclaimer.html. Time to ADAPT

Dynamic strategy, CLSA’s most popular participation rate algorithm, is empowered with ADAPTIVE’s short-term price prediction model. Dynamic alternates between two volume participation rates based upon not only a real-time benchmark but also a prediction benchmark anticipating where the stock price will go within the next few minutes. ADAPTIVE Benchmark works best in high volatility and strong momentum market conditions by predicting the target price.

For an optimal execution, ADAPTIVE Inline manages the overall participation rate on the order within trader defined aggression bands.

ADAPTIVE Inline calculates participation targets based on short-term predicted price, real-time volatility and price signals. The same price level at different times and market situations can result in a variable participation target. ADAPTIVE Inline trades up to the top of its range ahead of adverse moves, slowing down during the adverse, allowing for catch-up post reversion price moves.

ADAPTIVE BENCHMARK FOR DYNAMIC STRATEGY

ADAPTIVE Inline Algorithm

REAL-TIME VOLATILITY BAND

Participation Target Decreases

Participation Target Increases

Participation Target Catch up

Time

Price

22300

09:00 09:30 10:00 10:30 11:00

22250

22200

22150

ADAPTIVE Inline (Volume Inline)ADAPTIVE STRATEGIES

ADAPTIVE’s signal detection, price and volume prediction science puts Alpha into Inline strategies.

Compared to a pure volume reactive Inline strategy, the new ADAPTIVE Inline differentiates momentum and mean reversion scenarios, allowing the algorithm to adjust trading patterns to optimise execution performance.

May 2018 Time to ADAPT

Optimal participation rate is derived by balancing market impact cost estimation and D-VPIN adjusted timing risk. VPIN is a metric which measures the probability of informed trading in the market with a dimension of volume imbalance between buy and sell initiated trades. Based on VPIN, CLSA introduced the directional element to create D-VPIN as an adjustment indicator for timing risk estimate.

Real time volume prediction model further enhances the accuracy of market impact cost and timing risk estimation from the Implementation Shortfall model.

IMPLEMENTATION SHORTFALL MODEL

Market ImpactTiming Risk

Percentage of Volume (POV)

Timing risk adjusted with favourable D-VPIN Timing risk adjusted with neutral D-VPIN Timing risk adjusted with adverse D-VPIN Market Impact Cost Estimation using machine learning

bps

Optimal POV - Adverse informed market

Optimal POV - No significant information

Optimal POV - Favourable informed market

ADAPTIVE IS (Arrival Price)ADAPTIVE STRATEGIES

ADAPTIVE empowers the Implementation Shortfall model to optimise transaction costs by balancing market impact cost and timing risk. By continuously measuring the intraday volatility, D-VPIN (Directional Volume Synchronised Probability of Informed Trading Indicator) and volume prediction from Neural Networks, the Arrival Price strategy keeps adjusting the optimal participation rate in real time.

DISCLAIMER: This publication is subject to and incorporates the terms and conditions of use set out on www.clsa.com/disclaimer.html. Time to ADAPT

Price

100%

Volume(shares)

80

0

12:0

0:00

13:0

0:00

14:0

0:00

15:0

0:00

16:0

0:00

17:0

0:00

18:0

0:00

20

40

60

120000

80000

100000

0

12:0

0:00

13:0

0:00

14:0

0:00

15:0

0:00

16:0

0:00

17:0

0:00

18:0

0:00

20000

40000

60000

2820

Time

Time

2740

2760

2780

2800

2640

2660

2680

2700

2720

ADAPTIVE’s volume prediction model is powered by the neural network to provide an accurate prediction of closing volume in a sporadic liquidity environment. The model takes into account the day’s volume traded in the market as input to compare against historical volume profile for volume prediction and continuously evaluates the projection as the market trades in real time. The strategy execution plan is adjusted accordingly to allocate volume reserved for the close and find an optimal time to trade pre-close.

D-VPIN is applied on top of the volume prediction model, thus a buy and sell order on the same day maybe executed differently.

The upper chart shows the execution progress of a buy and sell QMOC order.

Backed-off and pushed more volume into the close in anticipation of high close volume as market volume improves

Buy done Sell done

EXECUTION PROGRESS

INTERVAL VOLUME AT TIME

ADAPTIVE market on close (QMOC)ADAPTIVE STRATEGIES

The QMOC strategy’s volume estimation model is refined with ADAPTIVE to provide a more timely and dynamic volume prediction.

The QMOC strategy may start early if predicted volume is low in the close and keeps revising the execution plan based on real time closing volume prediction.

Market Volume Price

May 2018 Time to ADAPT