CONTRIBUTORS Di Wu Vice President, Algorithmic Trading, ITG, Inc. di.wu@itg.com +1.212.588.4145 Wenjie Xu Quantitative Analyst, Algorithmic Trading, ITG, Inc. wenjie.xu@itg.com +1.212.588.4298
CONTACT Asia Pacific +852.2846.3500 Canada +1.416.874.0900 EMEA +44.20.7670.4000 United States +1.212.588.4000 info@itg.com www.itg.com
Algorithmic Portfolio Trading: A Primer Portfolio trading involves the execution of a basket of securities simultaneously. Compared to trading multiple securities independently, portfolio traders must consider a number of additional execution objectives in their decision-making process, such as minimizing risks, obtaining cash targets, and investing in multipleasset classes. For example, a portfolio strategy might require that securities in similar sectors remain hedged in order to reduce systematic risk; or have special funding needs to execute buys and sells in a dollar-neutral fashion; or involve equitizing cash by trading future contracts along side with equities. All of these complex and often conflicting objectives make portfolio trading inherently challenging. Developments in algorithmic trading play a critical role in addressing these challenges. Through sophisticated quantitative techniques, algorithmic strategies make it possible to incorporate major portfolio trading factors such as the composition of baskets, transaction cost of individual orders, volatility of the market, and price correlations amongst assets. These factors can be quantified and modeled under a unified optimization process to construct optimal execution strategies by minimizing transaction cost and portfolio risk simultaneously. Due to its complexity, algorithmic portfolio trading is often considered hard to understand and difficult to navigate. In this paper, we first provide an overview of a typical portfolio algorithm framework. We then discuss in detail the three key building blocks of this framework: strategic execution plan, tactical order placement, and real-time portfolio risk/cash management. We conclude with an explanation of the value that portfolio algorithms contribute to the overall investment process.
1.
PORTFOLIO ALGORITHMIC FRAMEWORK
In their seminal research, Almgren and Chriss (2001)1 laid down the theoretical foundation of optimal portfolio transactions. Years have passed since then; however, it still remains unclear to many practitioners how algorithmic strategies can achieve best execution through incorporation of a variety of trading objectives, constraints, and market conditions. To address this lack of transparency, we outline a portfolio algorithm framework that consists of three components:
Almgren R., and Chriss N., “Optimal Execution of Portfolio Transactions�, Journal of Risk, Winter 2000/2001. 1
2
¬¬ A Strategic Execution Plan generates an optimized trading schedule for each security by taking into account order sizes, expected transaction cost, estimated volatilities, correlations with other securities in a basket, and volume/cash constraints. ¬¬ Tactical Order Placement makes micro level decisions on how to best execute child orders according to the optimized plan. Specifically, for each security in a portfolio, algorithms choose the most appropriate trading tactics (i.e., passive vs. opportunistic) and trading venues (i.e., lit markets vs. dark pools) in order to minimize total trading costs. ¬¬ Real-time Portfolio Risk/Cash Management controls the trading process by reacting to changing information, such as risk exposures and cash imbalances. Once the control logic determines that execution plans and/or trading tactics have to be updated, it instructs the other two components to make adjustments accordingly. Figure 1 illustrates this framework. Note that it is a dynamic process in which all three components automatically adapt to real-time liquidity conditions and market movements. Figure 1. Portfolio Algorithm Framework
Source: ITG
2.
STRATEGIC EXECUTION PLAN
A strategic execution plan slices large orders into small tranches and schedules quantities of shares to be traded at different points in time. This can be achieved by constructing an “optimal” plan that has the least total trading cost. Just like human traders, algorithms balance realized transaction cost against residual portfolio risk2 in order to minimize total trading costs. A basket can be executed patiently with low transaction cost, but patient trading comes with the risk of future adverse market movement. In contrast, aggressive trading reduces risk, but incurs higher transaction cost due to increased market impact. This relationship can be illustrated by a total trading cost frontier in Figure 2.A. Depending on the degree of risk aversion, traders choose different urgencies (e.g., high vs. low) that correspond to various combinations of cost and risk on the frontier. Accordingly, algorithms generate execution plans with appropriate schedules (e.g., frontloaded vs. backloaded) to reach the desired level of risk exposure. Figure 2.B shows an example of two execution plans for a hypothetical portfolio.
Though total volatility is a commonly used proxy for the riskiness of portfolios, algorithms are flexible and can incorporate other risk measures such as cash risk, market risk, and sector risk. 2
3
Figure 2.A. Total Trading Cost Frontier
Figure 2.B. Different Risk, Different Execution
Source: ITG
In addition to incorporating urgencies, portfolio algorithms also utilize quantitative approaches to minimize risk, leading to better execution performance. For example, algorithms consider the correlations of securities when building execution plans. If a buy and a sell are highly correlated, both orders can be executed more slowly since one “naturally” hedges the other. Algorithms can also recognize that a buy/sell imbalanced basket has greater risk than a balanced one. As a result, the larger side of an imbalanced basket will be traded more quickly to reduce market/cash risk, all else equal. Furthermore, algorithms are capable of identifying the orders that impose significant risk to a portfolio, such as a large order in an illiquid stock without a “natural” hedge on the other side of the basket. A well-designed portfolio algorithm will try to execute such orders quickly if it is not too costly to do so. The final outcome of all these risk-minimizing strategies is a total trading cost efficient frontier, i.e., the regular frontier is optimized in such a way that transaction cost can be further reduced for a given level of risk (Figure 3.A). Then, based on the efficient frontier, portfolio algorithms can generate an execution plan that trades more patiently to achieve lower market impact without increasing risk (Figure 3.B).
4
Figure 3.A. Total Trading Cost Efficient Frontier
Figure 3.B. Same Risk, Different Execution
Source: ITG
3.
TACTICAL ORDER PLACEMENT
After an execution plan is determined, portfolio algorithms must choose appropriate order placement tactics to best execute the desired shares in the marketplace. Algorithms must be able to handle a basket of orders with varying characteristics such as size, spread, and volatility. In particular, two critical decisions have to be made: a) when to trade passively vs. opportunistically; and b) when to source liquidity from lit markets vs. dark pools. 3.1 Passive vs. Opportunistic The advantages of passive trading are obvious: spread capture and reduced market impact. Disadvantages include possible adverse selection and increased opportunity costs. Opportunistic trading, on the other hand, allows algorithms to aggressively take liquidity under favorable conditions, but at higher spread costs and market impact. As a result, portfolio algorithms constantly make tradeoffs between being passive and opportunistic.3
Bloomfield R. J., O’Hara M., and Saar G., “The “Make or Take” Decision in an Electronic Market: Evidence on the Evolution of Liquidity”, Journal of Financial Economics, 75.1, 2005. 3
5
In the context of trading a basket, it becomes more complex to make such decisions since the dynamics of order placement must be considered from the portfolio perspective. For example, a passive strategy may minimize transaction cost for individual orders, but is not necessarily suitable for a basket if executing a given order aggressively will reduce overall risk quickly. Based on quantitative measures of order characteristics (e.g., liquid stocks vs. illiquid stocks) and portfolio risk profiles (e.g., risky orders vs. non-risky orders), algorithms are able to determine an optimal combination of passive and opportunistic trading for each order. This ability and the capacity to instantaneously process market signals allow algorithms to improve the quality of order placement and, ultimately, execution. 3.2 Lit markets vs. Dark pools As dark pools continue to prove to be an important source of liquidity, algorithmic trading evolves apace with this market microstructure shift. Portfolio algorithms are no exception. While dark trading typically results in spread saving and low market impact, the probability of non-execution imposes a significant challenge in the portfolio context, because synchronized execution among securities is required and the risk of opportunity costs due to unfilled shares is high. One solution is to apply a base litmarket execution strategy and overlay it with an opportunistic dark-pool trading strategy. The former guarantees that trading progress can be made to follow strategic execution plans; the latter responds to market intelligence and takes advantage of favorable liquidity conditions with the goal of reducing transaction cost and/or risk. In addition to non-execution risk, dark trading is also subject to adverse selection. To overcome this problem, algorithms can adopt intelligent order placement logic to maximize fill rate while minimizing unfavorable post-trade price movement. Disciplined approaches, such as predicting execution quality by fill types among dark pools4, can significantly improve execution quality by avoiding interaction with the contra parties whose flow is often toxic.
4.
REAL-TIME PORTFOLIO RISK/CASH MANAGEMENT
Since trading is a dynamic process, any execution plan or order placement decision made at a prior point in time is subject to change later in response to new market dynamics as they emerge. Portfolio algorithms monitor execution progress in real time and adjust trading strategies continually. Among various execution measures, portfolio risk and cash position are two of the most critical that must be carefully managed. While lit-market and dark-pool trading take place, the portfolio risk of a basket fluctuates. For example, when the market rallies, sell orders are usually filled faster than the buys and cause dislocation in risk exposure. To overcome this issue, risk control periodically re-runs the risk minimization process (as described in Section 2) and updates the current execution plan. Figure 4.A shows a hypothetical scenario in which portfolio risk is managed for a dollar-neutral basket. When market goes up, sell orders fill faster, leading to an increase in overall portfolio risk. One possible solution for algorithms is to delay the execution of the sells but accelerate the buys. This will bring overall risk exposure back in line. Managing cash position is also important for trading portfolios, which often have mandates to reach specific cash targets and/or keep certain ratios between buy and sell values (e.g., dollar neutral). This is a constraint to be built into risk minimization, and the resulting execution plan usually maintains cash balances within tolerable
$
Polidore B. “Dark Pool DNA: Improving Dark Pool Assessment�, Journal of Trading, Spring 2012.
6
ranges. However, when liquidity “shocks” occur (e.g., a few large blocks are crossed almost simultaneously), a basket may suddenly become oversold or overbought. Under this circumstance, algorithms need to reduce the temporary imbalance by holding back the “fast” side, thereby allowing the “slow” side to catch up. Figure 4.B illustrates how cash balance is controlled for a hypothetical portfolio. Figure 4.A Managing Portfolio Risk
Figure 4.B Managing Cash Position
Source: ITG, Inc.
Note that well-designed portfolio algorithms do not solely rely on this real-time control logic to manage portfolio risk and cash position. A strategic execution plan is already in place to maintain desired levels of risk and cash throughout the entire trading horizon. However, when unpredicted events (such as market swings or fills from opportunistic trading) occur, real-time management is the most effective mechanism to keep portfolio risk and cash position in line.
5. CONCLUSION Portfolio traders are often confronted with complex decision making when executing portfolios. As the number of securities increases, it becomes progressively more challenging to manage the entire execution process. This paper discusses a portfolio algorithm framework with three components: a strategic execution plan, tactical order placement, and real-time portfolio risk/cash
7
management. We demonstrate that by leveraging quantitative approaches and advanced technologies, portfolio algorithms can play an important role in the pursuit of best execution. It is important for traders and portfolio managers to stay current with developments in algorithmic trading and identify the opportunities where portfolio algorithms can add value to the overall investment process.
Š 2012 Investment Technology Group, Inc. All rights reserved. Not to be reproduced or retransmitted without permission. 71912-21131 These materials are for informational purposes only, and are not intended to be used for trading or investment purposes or as an offer to sell or the solicitation of an offer to buy any security or financial product. The information contained herein has been taken from trade and statistical services and other sources we deem reliable but we do not represent that such information is accurate or complete and it should not be relied upon as such. No guarantee or warranty is made as to the reasonableness of the assumptions or the accuracy of the models or market data used by ITG. These materials do not provide any form of advice (investment, tax or legal).