INTRODUCING IS ZERO: Reinventing VWAP Algorithms to Minimize Implementation Shortfall

In a recent survey, we found that more than 72% of traders use the VWAP algorithm to minimize Implementation Shortfall (IS) for low-urgency trades (i.e., cost vs. midpoint price at order arrival). This preference underscores the limitations of traditional IS algorithms. Often branded as liquidity-seeking or arrival price algorithms, IS algorithms aim to balance execution risk against market impact. Yet, they often fall short for traders who prioritize long-term average cost efficiency over the variance of individual trades’ outcomes. This creates a paradox where VWAP becomes the default choice for those focusing on impact over execution risk, including quantitative portfolio managers who—due to high turnover and diversified portfolios—find execution risk less concerning than market impact.

VWAP algorithms spread orders throughout the day by design, minimizing participation rate and reducing market impact. In contrast, IS algorithms tend to front-load trades, either by design or in an attempt to opportunistically seek liquidity. Even at their lowest urgency settings, IS algorithms still place some emphasis on execution risk, naturally speeding up trading. While VWAP algorithms are currently the best available option to minimize average IS, they are not primarily designed for this purpose. Their main goal is to track the VWAP benchmark, and there’s much more an algorithm can do to minimize IS if not constrained by this benchmark.

To bridge this gap, we’re introducing IS Zero, a purpose-built algorithm that combines the desirable features of VWAP—such as non-urgent execution and distribution of orders throughout the day—with improvements on its deficiencies to better minimize IS performance. 

As the VWAP algorithm is the default strategy for these low-urgency trades aimed at minimizing IS, this paper outlines the deficiencies of the VWAP algorithm and how IS Zero addresses them. We also present empirical evidence showing how IS Zero improves average implementation shortfall performance compared to VWAP.

A Market Impact Minimization Schedule Differs from a Volume Curve

Since the inception of the volume-weighted average price (VWAP) algorithm, the prevailing belief has been that the best way to reduce market impact for a given order size is to trade in proportion to the market volume, effectively maintaining a constant participation rate. Indeed, trading in line with volume does mitigate market impact. However, participation rate, while correlated with the market impact an order creates, is not the only contributing factor—volatility and other measures play a crucial role.

The chart below illustrates the participation rate and empirical market impact for orders traded in a single 15-minute bin throughout the day, with the order size remaining constant¹. Trading with VWAP supports a fixed participation rate, but this approach is shown to be suboptimal. The chart confirms that higher volume can help reduce costs, as observed in the last few bins of the day when volume tends to be high and market impact is correspondingly low. However, this relationship doesn't always hold; at times when the same size order has a slightly larger participation rate in the middle of the day than at the beginning, costs can nearly triple, particularly early in the day—despite its high volume. The illustration below underscores that adhering strictly to a volume profile does not always minimize cost.

In contrast, IS Zero adopts a more holistic approach, utilizing a schedule that accounts for all factors influencing market impact, including volume and, notably, volatility–a significant factor according to our research. In scenarios where two bins have equal volume, the bin with lower volatility will typically have less market impact, and IS Zero prioritizes those bins.

The schedule for IS Zero, represented in blue for liquid stocks in the chart below, shows a clear departure from VWAP’s schedule as a result of the early volatility typically encountered in US equity markets just after the market opens. The corresponding VWAP schedule is depicted in navy for comparison.

In practice, IS Zero’s schedule allocates 21% of a full-day order over the first two hours of the day, while VWAP allocates 28%². IS Zero shifts the difference into the midday period, when prices are more stable. In the final hour of the day, IS Zero trades 25% of the order, in contrast to VWAP’s 34%. This shift in schedule improves IS performance on average, as detailed in later sections; however, it is important to note that IS Zero may incur higher tracking error to the VWAP benchmark as a result of this shift as well.

Dynamic Flexibility Makes IS Zero More Opportunistic

Implementation shortfall can be divided into components—costs due to market impact, paying spread, and adverse selection. While we've discussed market impact in earlier sections, the spread and adverse selection components also significantly affect performance. The spread component is often negative for VWAP orders below 5% of average daily volume, as VWAP algorithms typically provide liquidity. However, capturing this spread comes at the cost of adverse selection, as these algorithms are more likely to fill their passive orders when prices are about to become more favorable. 

Most schedule-based algorithms offer flexibility around the schedule to provide liquidity, capture spread, and take liquidity opportunistically. This flexibility is generally defined by "bands" around the schedule, allowing the algorithm to fill less or more than the scheduled amount. These bands are usually fixed, for example, allowing an algorithm to be 5 minutes ahead or behind schedule. The specifications are often independent of the product being traded, its liquidity, and evolving market conditions.

Increased flexibility in execution algorithms has several implications. First, it naturally increases the variance of performance against the underlying benchmark. For VWAP algorithms, this means a wider distribution of performance versus the interval VWAP benchmark. Adding flexibility around the schedule also permits more time for passive executions, reducing market impact and spread costs, particularly while trading long-queue or low-liquidity products, and lessens the need for aggressive orders during high-volatility and low-volume periods. However, higher flexibility also leads to increased adverse selection costs. For instance, if prices move in the trader's favor (e.g., prices declining in the case of a buy order), the algorithm will likely get ahead of its schedule, executing passive limit orders and orders resting in dark pools—both subject to adverse selection. Conversely, when prices move against the trader, the algorithm tends to lag behind. 

VWAP algorithms, aiming to reduce variance against the VWAP benchmark, often use inflexible bands regardless of the stock's liquidity. IS Zero does not focus on closely tracking VWAP; rather, it aims to minimize average spread costs and adverse selection costs, which are negatively correlated, and band flexibility is a variable that can help control these costs. Our research has determined optimal levels of flexibility by simulating different bands for various liquidity buckets and times of the day. We found that for stocks with less liquidity, wider bands are more effective as the reduction in spread costs outweighs the increase in adverse selection costs, thus improving net IS costs. For higher volume stocks, tighter bands than our default VWAP bands are more suitable due to the substantial adverse selection penalty associated with larger bands and the minimal benefits of lower spread costs, as average spreads for liquid stocks are typically low.

Our research also revealed that band sizes should vary throughout the day based on volume. Near the close, when volumes are high, bands should be tighter, and during low volume periods, wider.

Simulations across different liquidity buckets led to the optimization of these band sizes. The chart below presents a summary cost decomposition for US equities based on their daily volume. Nearly all groups show that dynamic bands significantly reduce slippage versus the arrival benchmark as both spread and adverse selection components shift.

Following simulation, these changes were implemented in IS Zero, and early A/B test results of this feature alone, including more than 2,500 parent orders indicate a reduction in cost versus arrival price from 8.8 basis points to 2.8 in this particular sample—a reduction of 68%—though variability increased slightly. Cost versus the VWAP benchmark also declined in this sample, though not the aim of this algorithm. 

IS Zero now utilizes this dynamic flexibility, which can adjust to be as much as 30 minutes ahead or behind schedule for long-duration orders of illiquid products, and as little as 1 minute ahead or behind schedule for liquid products.

High-Quality Dark Liquidity Reduces Market Impact

VWAP algorithms typically provide liquidity on exchanges and, to a large extent, in brokers' own dark pools. This approach stems from the fact that reducing market impact is not a primary concern for VWAP algorithms, and providing liquidity in dark pools does not necessarily reduce slippage against the VWAP benchmark. However, trading in Alternative Trading Systems (ATSs) offers the advantage of not displaying orders, thereby reducing market impact compared to displayed orders on exchanges. While hidden orders can be posted on exchanges, they are typically placed at the back of the queue due to price-display-type-time priority, leading to higher risk of adverse selection as they are more likely to execute when large market orders sweep the price level.

Displaying orders on exchanges is also beneficial for algo providers, as it generates higher rebates. In fact, VWAP algorithms are likely the most profitable for providers in terms of rebate generation, especially when the algo provides liquidity exclusively in the provider’s own dark pools, thereby generating additional commissions by attracting the other side. This behavior aligns with the primary objective of VWAP algorithms, which is to minimize cost versus the VWAP price. While this aligns with the stated objective of VWAP algorithms and providers’ own interests, it clearly does not align with traders’ intended goal when utilizing VWAP algorithms to minimize IS costs. 

IS Zero, with its stated objective of minimizing Implementation Shortfall, optimizes liquidity access across exchanges and ATSs. We have found that accessing liquidity in dark pools does not typically reduce cost versus the VWAP benchmark—except for orders representing a large percentage of average daily volume (ADV)—but it does effectively reduce IS costs. Skillful management of liquidity in dark pools is crucial due to their more opaque and complex market structure than exchanges’. Our approach to liquidity curation in dark pools involves a combination of opt-in/opt-out, optimizing the segmentation offered by ATSs, setting minimum order sizes, applying pegging instructions, and using an optimal mix of posting and pinging. While we conduct extensive liquidity curation processes and controlled experiments to assess the effectiveness of our methodology, a detailed discussion on this topic is beyond the scope of this paper and may be addressed in future work.

What is pertinent to this paper is that IS Zero utilizes our liquidity curation process and optimally balances dark liquidity with exchange liquidity while adhering to the impact-optimized IS Zero schedule described above. This approach has been shown to improve IS performance, although it does not necessarily reduce slippage versus the VWAP benchmark. Example venue statistics are shown below for orders traded in our VWAP strategy versus IS Zero, highlighting the preference of IS Zero for dark execution when available.

The Results: IS Zero vs. VWAP Performance Comparison

IS Zero's design is demonstrably enhancing performance for our clients. It not only outperforms our VWAP algorithm in terms of implementation shortfall for low-urgency orders but also exceeds in performance against its own interval VWAP benchmark.

Comparing the performance of two algorithms is a complex task. Biased and/or unreliable results can arise if the samples of orders traded are not sufficiently large, face differing market conditions, or vary in order difficulty. To mitigate sampling bias, randomized sampling is the most effective method. An ideal comparative study involves randomly routing parent orders to the strategies under comparison. With a large sample size, we expect random samples to be statistically similar in characteristics such as market conditions, order duration, spread, size, participation rate, and liquidity characteristics.

Strategy Studio, a component of our Algorithm Management System (AMS), facilitates the quick creation of A/B tests as described above, by randomly allocating parent orders across execution strategies in a specified proportion. We employed Strategy Studio's A/B testing functionality to compare IS Zero with VWAP using a sample of low-urgency orders from clients interested in the strategy. Over a three-month period, 144,000 parent orders routed to both strategies exhibited similar characteristics in terms of order difficulty, including participation rate and duration.

According to the table, IS Zero's implementation shortfall is 13.1 basis points, compared to 20.8 basis points for VWAP, marking an average improvement of 37% in this sample.

Additionally, examining order flow from a price impact perspective further highlights IS Zero’s reduction in market impact. The chart below illustrates the intraday average realized implementation shortfall (versus the arrival price) of orders sent to our VWAP algorithm (navy) and IS Zero (blue) as they are executing, measured during the period of A/B testing. VWAP orders exhibit the expected square-root relationship between time and incurred market impact. IS Zero’s drift represents a steep departure from VWAP’s impact curve, with its orders creating substantially less impact both at the onset of the order and beyond.

IS Zero’s updated schedule, dynamic flexibility, and increased dark execution are helping it outperform a VWAP strategy for orders with low urgency (no expected alpha). IS Zero from BestEx Research is a more targeted way to trade low-urgency orders to reduce implementation shortfall from all angles, tackling the market impact, adverse selection, and spread costs that contribute to improved IS performance.  

Moreover, IS Zero provides us with a robust platform for ongoing research and improvement in implementation shortfall performance for clients who prefer to distribute their trades over the entire trading horizon without a strong focus on closely tracking the VWAP benchmark. In this way, IS Zero not only delivers immediate performance benefits but also paves the way for continuous advancements in trading strategy optimization.

‍

‍

‍

‍

End Notes

¹The illustration shown here represents orders sized at or below 0.5% of the stock’s average daily volume for Russell 3000 constituents with spreads at or below 5 basis points. However, charts for all order sizes and spread groups showed exceedingly similar patterns though scales differed.

²This chart shows average schedules for Russell 1000 constituent stocks for illustration purposes.

‍

‍

At BestEx Research, we care how you fill. We know from experience that systematic, quantitative decision-making around algorithm design contributes to globally optimal execution and results in significantly reduced execution costs. Reach out to us with questions at research@bestexresearch.com or learn more about us at bestexresearch.com. 

This research paper reflects the views and opinions of BestEx Research Group LLC. It does not constitute legal, tax, investment, financial, or other professional advice. Nothing contained herein constitutes a solicitation, recommendation, endorsement, or offer to buy or sell securities, futures, or other financial instruments or to engage in financial strategies which may include algorithms. This material may not be a comprehensive or complete statement of the matters discussed herein. Nothing in this paper is a guarantee or assurance that any particular algorithmic solution fits you, or that you will benefit from it. You should consider whether our research is suitable for your particular circumstances and needs and, if appropriate, seek professional advice.

©2024 BestEx Research. All rights reserved.