Taming the "Beast": TCA on QMIT indices

Sendoh Xu & Brian MacSweeney [with guidance from Milind Sharma]*

• Both authors are MSCF candidates at Carnegie Mellon and would like to thank Matvei Lukianov for editorial input.

• Special thanks to Chris Sparrow of Factset for providing the BEAST model data & valuable feedback.

Introduction and Context

Efficient trade execution is paramount for quantitative investors and portfolio managers. Even with a sound investment strategy, excessive transaction costs can erode returns. To uncover what drives these costs, we analyzed a comprehensive dataset of estimated trading costs across approximately 2,600 U.S. equities in QMIT universe on June 30, 2025. Using FactSet’s BEAST transaction cost model, we simulated a standardized trade size ($1 million executed between 09:30 AM and 4:00 PM) for each stock and examined the expected cost per share (in basis points) and its components excluding the MOO & MOC auctions at the open & close. This cost per share (in basis points) represents the model’s estimate of implementation shortfall – the slippage between the trade’s decision price (aka arrival price) and the average execution price. It encapsulates both spread and liquidity costs. In this article, we explore our analysis of liquidity measures, examining how trading volume and market capital influence trading costs. Our goal is to distill practical insights into the drivers of execution costs and strategies traders can use to mitigate them. Key questions include: How are trading costs distributed across stocks? What roles do volume, liquidity, and volatility play? What can practitioners do to minimize the drag of transaction costs?

The Landscape of Trading Costs

Definition

The FactSet BEAST transaction cost model decomposes total implementation shortfall (measured as cost per share in basis points) into clearly defined components, following principles derived from market microstructure and optimal trade execution theory:

Total Cost = Spread Cost + Liquidity Cost

• Spread Cost: Represents the explicit cost of crossing half the bid-ask spread. This cost is deterministic and directly linked to market liquidity at the moment of execution.

• Liquidity Cost: Reflects the temporary market impact (implicit cost) due to consuming available liquidity—“eating the book”—during order execution.

• Market Impact: Separately reported but NOT included in the total transaction cost, market impact captures the permanent shift in the mid-price from the order's start to its conclusion. This differs from total cost because expected cost compares the avg price with the arrival price. If you start buying at $10 and move the price up to $11 at the end of your order, with an avg price of 10.51, then the cost is .51 and the market impact is $1.

By separately reporting these distinct cost components, the BEAST model enables traders to better identify the sources of execution slippage and implement targeted strategies to minimize transaction costs.

Notes on Scope and Cost Definition:

• Implementation Shortfall ('Cost' or 'Slippage') refers to Total Cost, i.e., spread + liquidity, though often conflated with generally-speaking “market impact”.

• All estimated transaction costs presented here explicitly exclude brokerage commissions and other explicit fees, focusing solely on implicit trading costs (spread, liquidity, etc.) as defined by the BEAST model.

• The cost estimates and analyses provided in this write-up includes liquidity available on lit markets (primary exchanges), and from Alternative Trading Systems (ATS), dark pools, or other off-exchange venues.

Cross-sectional t-cost profile for QMIT universe (June 30, 2025)

Table 1: Summary of Cost BPs (based on $ 1mil trade)

Figure 1: Distribution of costs (bps) across the universe

Our analysis revealed trading costs are highly skewed across stocks. For the majority of stocks, predicted costs are minimal – a few basis points or less – reflecting little more than the cost of crossing the bid-ask spread. However, a small subset of stocks incurs extremely high costs, creating a pronounced long tail in the distribution. While the vast majority of stocks can be traded with negligible cost, the presence of these outliers introduces significant tail risk to transaction costs.

In our one-day sample, at the 90th percentile, costs rise sharply, and only in this upper decile of highest-cost names do we observe execution costs climbing into the dozens or even hundreds of basis points. In the most extreme cases, the model predicted costs exceeding 1000 basis points for certain illiquid stocks implying that a single trade could move the price by more than 10%. Though infrequent, these extreme outcomes exert a disproportionate influence on the overall distribution of trading costs. In the decile histograms below, we use the right end point (max) of the interval. Costs are reported in dollars per share or in basis points (where 1bp is 0.01%).

Figure 2: Distribution of Cost per share by decile

Unsurprisingly, the low-cost group is dominated by highly liquid, large-cap stocks. Mega-cap names like Apple, Amazon, and NVIDIA often incur virtually no market impact for a trade of $1 million.

These stocks trade tens of millions of shares daily, so a $1 million order hardly dents the order book. In our data, 550 stocks show no liquidity cost at all but on average show 5.3 bps of t-cost which is substantial. For these names, trading costs are limited to the bid-ask spread, which is typically just a few basis points in highly liquid markets

On the other hand, the high-cost tail consists almost entirely of low-volume, small-cap stocks where even modest trades can push deeply into the order book and generate significant costs. Poor liquidity in these names means that sizable orders can easily move prices, creating a stark divide in trading costs across the universe of stocks.

As a result, while the majority of a typical portfolio can be traded at minimal cost, a small number of illiquid positions can disproportionately drive up overall transaction costs. This skew suggests that trading cost budgets are often dictated by a handful of problematic names rather than the broader portfolio. Even otherwise cost-efficient strategies can be significantly distorted by just one or two expensive-to-trade stocks, making it essential to proactively identify and manage these outliers.

Anatomy of Trading Costs: Spread vs. Liquidity

Every trade’s implementation cost consists of two parts: spread cost and liquidity cost. Spread cost reflects the immediate expense of crossing the bid-ask spread – typically half the spread if using the mid-price as a benchmark. Liquidity cost captures slippage from moving the mid-price, driven by order book depth and other price dynamics.

Spread cost is essentially a fixed cost per share for a given stock and time, while liquidity cost grows with the order’s size relative to available liquidity. Model outputs display a near-perfect correlation (~0.98) between total cost (in bps) and the liquidity cost component, and a strong correlation (~0.78) with the spread component. This result further confirms that trading costs are largely determined by these two factors putting aside Momentum costs which are not considered herein.

Furthermore, spread costs and liquidity costs are strongly positively correlated (ρ ≈ 0.64) amongst themselves. Stocks that exhibit high market impact often also have wider bid-ask spreads, reflecting their shared illiquidity characteristics. Thinly traded stocks typically feature both wide spreads and shallow order books, which can be quickly depleted by even moderately sized trades.

Table 2-1: Correlation Matrix (Pearson)

Table 2-2: Correlation Matrix (Spearman)

These cost components behave across firm-size and liquidity strata. Large-cap, actively traded stocks tend to have very tight bid-ask spreads (often 1–2 bps) and deep liquidity, while smaller or less liquid stocks tend to show wider spreads and shallower order books. Spread cost serves as a baseline cost for any trade—relatively stable and low for large caps, but significantly higher for small- and micro-cap names. In our universe, major stocks exhibited spread costs in the range of 1–5 bps, whereas some smaller names showed spreads of several tens of basis points.

Liquidity cost becomes significant when an order starts to exhaust the order book. For highly liquid stocks (e.g. top-decile volume) a $1 million order barely penetrated the first few levels of the book, causing minimal market impact. In contrast, for the least liquid names, the same order could represent multiple days of average trading volume, quickly consuming available liquidity and triggering large price swings. In these cases, liquidity dominates total trading cost.

This pattern is evident when analyzing cost component breakdowns by liquidity decile. In the lowest trading-volume decile, most trading cost comes from liquidity cost, with spreads contributing to only a small share of the cost. Pie charts for these stocks are heavily weighted toward liquidity cost (deciles 1 & 2). This indicates that trades are expensive mainly because the market cannot absorb size without significant price disruption. For larger cap names (deciles 9 & 10), however, orders of the same amount have a negligible impact on liquidity, so we observe that spread costs account for nearly all transaction costs in this scenario.

Figure 3: Cost Component Breakdown.

The Dominant Driver of Cost

Perhaps the single most important insight from the analysis is the critical role of liquidity in explaining variability in trading costs. Traditional liquidity measures such as market capitalization and 30-day average daily trading volume (ADTV) show a clear relationship with trading costs.

Figure 4: Scatter plot and OLS Fit (Log10 Cost vs Log10 Market Cap)

Figure 5: Scatter plot and OLS-fitted Line (Log10 Cost vs Log10 ADTV)

Intuitively, a strong inverse relationship exists between a stock's liquidity and its trading cost. Securities with higher average daily trading volume (ADTV) and larger market capitalizations have deeper order books and tighter bid-ask spreads. Consequently, executing trades in these stocks is cheaper as they can absorb larger orders with less price impact. This fundamental relationship is confirmed when examining rank-order correlation or grouped comparisons, the relationship strengthens considerably. Ranking stocks by market cap or volume reveals that trading costs decline consistently as liquidity increases. Scatterplots of cost versus market cap or ADTV show a clear downward trend: as stocks become larger and more liquid, total cost decreases. A simple trend line fitted to log10 data suggests that ADTV alone explains roughly 60% of the variation in cost ranks which is slightly more predictive than market cap. In short, ADTV emerges as one of the best predictors of trading cost. Stocks with low ADTV tend to incur higher-than-average costs, while those with high ADTV are typically low-cost trades. 

Grouping stocks into deciles by liquidity illustrates this pattern. The lowest-liquidity decile (by ADTV) had a much higher median transaction cost and a long tail of extreme (higher-cost) outliers, with many trades incurring dozens or even hundreds of basis points in costs very long upper tail of even higher-cost outliers. In contrast, as liquidity increases across deciles, the entire cost distribution shifts lower and tightens. By the highest-volume decile, median costs were near zero and high-cost outliers were rare. A similar pattern holds when grouping by market cap: trades in small-cap stocks are not only more expensive on average but also far less predictable, showing high variance and occasional spikes to punitive cost levels.

These findings highlight liquidity is the dominant force in determining trading costs.

Figure 6: Boxplot of transaction costs (bps) by liquidity group (deciles of ADTV).

From a practical standpoint, average daily volume is a direct proxy for how easily a stock can be traded. High-ADTV stocks provide consistent, low-cost execution strategies – a significant advantage for strategies that trade frequently or in size. Conversely, strategies that venture into low-volume stocks in search of higher alpha must be prepared for sporadic, but at times substantial trading costs.

The practical implication is clear: maintaining sufficient liquidity in portfolio holdings is essential for keeping transaction costs low and predictable. In practice, many institutional investors manage this risk by filtering or weighting investments based on average daily trading volume (ADTV), often applying minimum liquidity thresholds—such as excluding stocks that trade below a certain dollar volume per day.

Our data supports this approach. Simply avoiding the least liquid names eliminates the most extreme cost outliers and leads to a meaningful reduction in overall execution costs. ADTV-based filtering serves as an effective and scalable tool for managing transaction cost risk.

Market capitalization, which often correlates with volume, can also serve as a useful guide – large-cap stocks generally cost less to trade. However, size alone isn’t a perfect proxy for liquidity. Many mid-cap stocks trade heavily and are inexpensive to execute, while some large-cap stocks may trade less frequently and incur meaningful costs.

The analysis showed market cap and ADTV are strongly correlated (Pearson ρ ≈ 0.72 in our sample). Size alone is not a reliable predictor of trading costs. Direct liquidity measures such as bid-ask spread or ADTV are more actionable. For example, there were cases of mid-sized companies with high turnover that were cheap to trade, alongside some large but thinly traded stocks that carried non-trivial costs.

Figure 7: Scatterplot of cost per share (bps) vs. market cap. Each gray point is a stock. The few red points indicate “high-cost outliers”

Benchmark Index Cost Analysis

Figure 8: Weighted average cost of Indices

We now compare five indices {Dow 30, S&P 500, LBO 100, QUMN and the full “Universe”} under equal‐weighted and cap‐weighted schemes, decomposing total cost into spread and liquidity (Figure 8).

The QMIT LBO Top 100 Index (LBO100) tracks the alpha of an equally weighted market neutral portfolio whose holdings are approximately the Top 100 machine learning based leveraged buyout (LBO) candidates hedged by a dollar neutral short position in the IWN ETF (Russell 2000 Value). The index is rebalanced and reconstituted weekly.

The QMIT Equity Market Neutral Index (QUMN) seeks to track the performance of QMIT’s EMN (Equity Market Neutral) hedge fund model. The model creates market neutral long/short portfolios based on a universe of the largest approximately 2,500 US + dually listed Canadian stocks + ADRs. The underlying long and short portfolios leverage multi-factor alpha signals employing 18 ESBs (Enhanced Smart Betas) which are distilled from hundreds of factors via Machine Learning Ensemble Methods. The index is rebalanced and reconstituted weekly.

Under equal weighting, the Dow 30 remains the cheapest at 2.31 bps of total cost (2.31 bps spread, 0 bps liquidity), reflecting its deep, liquid constituents. The S&P 500 jumps to 5.01 bps total cost (4.94 bps spread, 0.07 bps liquidity), given the inclusion of less liquid names as we go further down the cap spectrum. The LBO 100 - comprised of the top 100 deep value small cap leverage buyout targets - registers 17.12 bps (11.06 bps spread, 6.06 bps liquidity), underscoring the outsized drag from liquidity. The QUMN index on the other hand registers 15.85 bps (11.58 bps spread, 4.26 bps liquidity). This dollar and beta neutral index monetizes our flagship composite alphas across the full Universe which is why it’s quite as high. The full Universe is much higher cost averaging 24.72 bps (13.28 bps spread, 11.44 bps liquidity), driven by the long tail of illiquid, small-cap names outside of the S&P 500.

With cap weighting, costs compress significantly. The Dow 30 falls to 1.19 bps (1.19 bps spread, 0 bps liquidity) by emphasizing its most liquid Mega-caps. The S&P 500’s total halves to 2.92 bps (2.92 bps spread, 0 bps liquidity), while the LBO 100 drops to 7.83 bps (6.45 bps spread, 1.38 bps liquidity) and QUMN drops to 6.17 bps(5.34 bps spread, 0.83 bps liquidity)—illustrating how a few larger-cap LBO names can materially reduce average cost. Even the full Universe cost shrinks to 4.19 bps (3.81 bps spread, 0.38 bps liquidity), as the largest, most liquid stocks dominate the index weighting.

This plot highlights two key insights: The market capitalization is extremely lopsided. When you do cap weighting at the universe level, it's going to dramatically lower the liquidity cost. Second, decomposition-wise, liquidity impact overwhelmingly drives cost in broad or small-cap‐heavy universes, and cap‐weighted exposures recapture most of the liquidity premium by tilting toward the largest, most tradable names.

Analysis via BEAST GUI

To complement our cross-sectional index comparisons, we also pulled single-stock analytics directly from the BEAST GUI (Date: 2025-06-30, Time: 9:30-16:00, Amount: 1m dollar), disabling momentum to isolate the core cost drivers and then comparing a high-liquidity large-cap name (Apple) versus a small-cap (AMWD, one of our LBO100 constituents).

Figure 9: Cost Estimate of APPLE in the BEAST GUI

Figure 10: Cost Estimate of AMWD in the BEAST GUI

For Apple (AAPL-US), the GUI reports a total implementation shortfall of 0.49 bps, with the orange total-cost line perfectly overlaid by the blue spread-cost line, indicating that spread alone accounts for the entirety of expected slippage and that liquidity impact, signaling, and momentum costs are all effectively zero. In other words, trading 1 million dollars of Apple under this scenario incurs only the cost of crossing the tight bid-ask spread.

By contrast, for AMWD-US the GUI estimates a 45.87 bps total cost (≈$0.2378 per share). Here the breakdown reveals a dominant spread component (≈41 bps) and a smaller but meaningful liquidity component (≈4 bps), with signaling and momentum costs still negligible when momentum is turned off. This stark divergence between the tiny, spread-only cost for a Mega-cap versus the large, mixed spread + liquidity cost for a small-cap, visually underscores how liquidity profiles drive implementation shortfall. Including these selected GUI screenshots in our write-up vividly illustrates the contrasting cost decomposition across capital-size regimes.

Managing the High-Cost Outliers

Identifying the handful of high-cost outliers is only half the battle – the next step is managing them. For a quant portfolio, trading cost outliers pose a risk to performance and need special attention. Our analysis suggests a couple of practical approaches to tame these cost drivers:

• Flag and potentially exclude extremely costly names: We set a threshold at the 90th percentile of cost (and impact) and defined any stock above that as a “high-cost outlier”. This cutoff effectively isolated the bottom ~10% of names that contributed a disproportionate share of aggregate trading cost. The implication is that by excluding the top decile of costliest stocks, one could eliminate the bulk of extreme transaction cost events, leading to a far more predictable cost profile. In fact, our study found that removing those outliers would significantly reduce the tail risk of transaction costs and yield a more controlled execution cost distribution. A portfolio manager might decide to avoid these tickers entirely, or replace them with more liquid substitutes, unless their return contribution is so high that it justifies the cost.

• Prefer large-cap, high-liquidity names: Since the outliers were predominantly low-ADTV, small-cap stocks, a portfolio skewed toward larger, more liquid equities will naturally have much lower transaction costs. For example, an equity long-short strategy might choose to confine its universe to the Russell 1000 (large/mid-cap) rather than the entire Russell 3000, as a way to control trading frictions. The trade-off is potentially fewer alpha opportunities outside the most liquid names. But our data clearly shows the execution cost benefits of focusing on liquid names – portfolios of liquid stocks exhibit tight spreads and minimal impact, lowering overall cost.

• Mitigate costs for necessary illiquid positions: There may be cases where high-cost names do appear in the portfolio – perhaps a niche small-cap stock with exceptional alpha potential that passes your filters. In those instances, trading should be handled with extra care. One might use smaller position sizes or trade in smaller batches over a longer time to reduce impact. Another tactic is to rebalance less frequently on those names (since each trade is expensive) – effectively accepting a bit more tracking error to avoid repeated cost. The key is to not treat a low-liquidity stock the same way as a blue-chip in trading plans. If you must engage with some higher-cost stocks, plan for it: employ liquidity-seeking algorithms, work the order passively if possible, and anticipate materially higher costs (and uncertainty) in your trading schedule. In short, adjust your execution strategy for the assets’ liquidity.

• Review and adjust the universe periodically: If you do exclude the worst offenders, it’s prudent to recompute your cost metrics after those exclusions to quantify the improvement. We found that dropping the top decile outliers dramatically improved portfolio-average cost statistics (e.g. the new 90th percentile cost might drop to what was previously the 70th percentile). Regularly monitoring which names fall into the high-cost bracket (since liquidity can change over time) and updating the exclusion list or trading protocol is a good practice. This before-and-after analysis provides tangible evidence of cost efficiency gains, which is valuable for strategy evaluations and for communicating to stakeholders that trading costs are under control.

It’s also insightful to examine the characteristics of the high-cost group as a whole. Besides low liquidity, some common themes emerged. Many of these names were small-cap stocks, often with high volatility and wide spreads, as expected. Some were companies going through distress or special situations (which typically reduces liquidity). We also observed the dispersion within the outlier group itself:: many had cost estimates on the order of 50–150 bps (high, but perhaps manageable for a big alpha); a subset was far worse, in the several-hundred bps range. This suggests that one could implement an additional tier of risk control – for example, setting an even higher cost cutoff (e.g. the 99th percentile) to identify the few worst offenders among outliers and eliminate those as well. The overarching point is that a small number of stocks can dominate your execution cost risk, so you should know who they are and have a plan for them.