The buy-side continues to face challenges in appropriately bucketing orders based on their alpha profiles and characteristics such as volatility, spread, and liquidity. Traditionally, algo wheels assign strategies based on metrics like average daily volume (ADV), notional, or by benchmark aiming to minimise trader bias.
In recent years, a shift toward arrival based, liquidity-seeking strategies has emerged, freeing traders to focus on larger, more complex orders.
Using Berenberg’s proprietary Trading Intelligence Analytics (TIA) platform, we analysed two years of trades across 25+ markets, incorporating historical and Level 3 market data to assess whether this shift resulted in better performance across all buckets of flow.
Should timing risk versus market impact drive strategy selection?
The short answer is, yes. Balancing timing risk (adverse price movements over the trading period) against market impact (price disruption from aggressive trading) is critical for algo selection in single-stock and program trading. This balance directly affects performance relative to the arrival benchmark. When choosing between a VWAP OTD strategy versus a liquidity seeking approach, traders assess whether the market impact of trading at higher participation rates outweighs the timing risk of slower execution.
“Our analysis shows that for certain order subsets, combining a VWAP strategy with an Implementation Shortfall (IS) overlay – termed VWAP-Arrival – yields significantly better execution outcomes against an arrival benchmark,” says Jason Rand, global head electronic trading and distribution at Berenberg. This is because the reduced impact from trading slower results in overall better execution, as opposed to a more aggressive strategy.
The evolving landscape: Declining lit market share
European equity markets have seen a steady decline in lit market volume share over the past seven years, with 2020 as a temporary outlier due to Covid-driven volatility. As lit order books shrink, optimal participation rates must adapt. Passive strategies increasingly outperform aggressive ones, particularly in low alpha decay or less time-sensitive environments. The proliferation of bilateral liquidity arrangements and their less understood impact on the lit order book could be another factor.
With that said, strategy selection remains nuanced. Key factors of consideration such as alpha decay (more critical for hedge funds than fundamental investors), ADV, spread, volatility, and time-of-day all influence whether an order should be executed passively or aggressively.
Understanding ADV and its limitations
The relationship between order size and ADV is foundational for setting participation rates. Larger orders relative to ADV require lower participation to minimise impact. However, relying solely on ADV can be misleading, as it includes full-day volume, penalising orders placed later in the day. Additionally, ADV incorporates closing auction volume, which can account for a significant portion of daily activity but does not reflect continuous trading conditions.
For example, an order to buy one million shares of VOD LN may seem modest at 2% of a 50 million share ADV. However, if placed at 2pm after 20 million shares have traded, the remaining daily volume (RDV) is 30 million, making the order 3.3% of RDV. If 30% of volume occurs in the closing auction, the effective participation rate in continuous trading rises to 6.7% – over three times the nominal ADV figure. Our VWAP-Arrival strategy dynamically adjusts to these real-time conditions, minimising slippage against the arrival benchmark.
Minimising adverse selection through adaptive logic
VWAP-Arrival mitigates adverse selection by evaluating quote stability, bid-ask dynamics, and real-time liquidity signals to detect informed flow. If risk is detected, the algo delays or repositions passive orders to avoid being ‘picked off’. This adaptability allows VWAP-Arrival to maintain passive execution in stable conditions while seeking price improvement without exposure to toxic liquidity, optimising the trade-off between execution risk and market impact. In scenarios where completion in the dark is necessary, there are two primary approaches available.
The first approach is employing a Relative Dark Limit (arrival, iVWAP, index/sector etc.). This operates as an “I WOULD” order, posting in dark or block venues at a specified improvement over a benchmark (e.g. 5bps better than arrival). It allows for passive interaction without crossing the spread, while still targeting price improvement.
The second approach is residual slicing by volume profile where the remaining order can be sliced and executed in the dark following the stock’s intraday volume schedule. This maintains alignment with natural liquidity patterns while attempting to minimise signalling risk.
Volume prediction accuracy as a competitive edge
Outperforming VWAP and arrival benchmarks hinges on accurate volume forecasts. Our data science team has enhanced our prediction models using feature engineering to improve accuracy.
One example is our Multi-Model Prediction Framework. Instead of relying on a 21-day average volume (filtered for outliers like rebalances or earnings), stocks are bucketed by sector, market cap, and primary market.
Forecasts are generated using historical means (5d, 10d, 21d), medians, and ARMA models (autoregressive moving average) and measured for accuracy. Daily, the engine selects the best performing model per stock based on recent prediction accuracy, improving reliability across all liquidity profiles.
The second example relates to day-of-week effects. Volume patterns vary by day but often persist week to week. Thursdays and Fridays often see higher volumes (15-20%) than Mondays and Tuesdays for example due to position unwinding, economic releases, or earnings. Incorporating these patterns enhances forecast accuracy.
Measuring what matters
To achieve consistent outperformance, algorithms need to adapt to shifting market regimes, liquidity conditions, and order routing biases while continuously back-testing predictions for accuracy.
In a back-test of over 2,500 single stock orders from 2023–2025, VWAP-Arrival improved median spread-adjusted arrival slippage by 4.9 bps compared to a market-adjusted benchmark. Unlike traditional VWAP, Berenberg’s VWAP-Arrival dynamically optimises for favourable trading conditions rather than adhering to a strict volume profile. It adjusts participation rates based on realised volatility taking into account implicit costs and fair value models, prioritising price dislocation over passive posting or spread capture.
To evaluate performance, we employ machine learning models to normalise execution quality by trade difficulty and market conditions. A more intuitive approach also involves comparing arrival-mid slippage versus interval VWAP (iVWAP). A strategy may beat the arrival benchmark in isolation but underperform iVWAP, indicating market relative underperformance rather than strategy success.
“Employing a scorecard methodology is often the most effective way to evaluate this performance as it combines multiple benchmarks (applying weight to each) to provide a comprehensive assessment of execution quality,” adds Olatayo Balogun, associate director, electronic trading at Berenberg.
