Can Supply-Chain Alpha Survive Transaction Costs?
September 14, 2026
Altsets
Research by Altsets Research
Relationship features can change slowly while portfolio rules create unnecessary turnover, so a tradable supply-chain strategy needs liquidity-aware costs, sensible rebalance frequency, and hold rules that preserve information without constantly trading noise.
Data used:Altsets Supply Chain Intelligence: 90k+ entities, 400k+ relationships, 20+ years of history.
Key findings
- Slow-moving relationship features do not require fast trading, and a high-turnover portfolio built from a persistent network signal can destroy net performance without adding much new information.
- Transaction-cost research shows that anomaly profitability can fall materially after costs and that wider hold regions can improve implementability, making turnover control part of signal design rather than an afterthought.
It can, but only if net performance remains positive after realistic trading costs. A supply-chain strategy that works only before commissions, bid-ask spreads, slippage, market impact, and turnover has not produced tradable alpha.
Slow relationship features can still create fast portfolios
Many supply-chain relationships are structurally persistent. The supplied Shin-Etsu Chemical view maps Samsung Electronics, TSMC, and Intel as customers with displayed supplier revenue shares of 2.43%, 4.02%, and 1.79% respectively. Those relationships do not imply that a quant should rebalance a portfolio every day. If the research hypothesis is based on customer concentration or dependency structure, a monthly point-in-time feature may contain the relevant information while a daily rebalance only creates more trades around a slowly changing signal.
The problem is that the trading rule can create turnover even when the feature itself barely moves. A rank-based long-short strategy can flip positions because two companies swap places near a portfolio cutoff, while an optimizer can make large trades after tiny forecast changes if no turnover penalty is present. Novy-Marx and Velikov show why this matters across a broad set of anomalies: transaction costs materially reduce profitability, and high-turnover strategies are much harder to preserve net of costs. Their evidence also shows that simple cost-mitigation rules, including wider buy-versus-hold thresholds, can improve implementability.
Measure turnover caused by the portfolio rule, not only feature turnover
A supply-chain feature can have a low raw change rate and still produce a high portfolio turnover rate after cross-sectional normalization, neutralization, and position sizing. The quant should therefore log several quantities separately: how often the relationship feature changes, how often a stock crosses an entry threshold, how often an existing position leaves the hold region, and how much notional value is actually traded at each rebalance. Looking only at signal autocorrelation can hide the execution burden created by the final portfolio construction step.
A useful experiment can compare several rebalance frequencies under the same point-in-time signal. If a monthly relationship feature produces nearly the same gross information coefficient at weekly and monthly rebalancing, but weekly turnover is much higher, the faster implementation is difficult to justify. The same logic applies to an event-conditioned signal. A customer earnings event may warrant rapid temporary trading, while a structural dependency score may be better used as a slow-moving portfolio constraint. The economic role of the feature should determine the natural trading frequency.
Cost assumptions should vary by stock and strategy capacity
Applying one constant cost to every trade is convenient and often unrealistic. Smaller and less liquid securities can have wider spreads and larger market impact than mega-cap stocks, while a market-neutral supplier basket can require shorting names with different borrow conditions. A serious backtest should at least make costs responsive to liquidity and traded size, even if the execution model remains simple. The objective is not to simulate every order-book event perfectly. It is to make sure the strategy does not depend on fills that become implausible when scaled.
Supply-chain strategies can create concentrated trading around the same event. If a customer surprise causes several connected suppliers to enter the model simultaneously, the portfolio may trade a cluster of correlated names at once. That can increase liquidity demand precisely when spreads and volatility are elevated. A transaction-cost model based only on calm-period averages can therefore understate the cost of the event-driven version of the strategy. Stressing costs during high-volatility or event windows can reveal whether the claimed alpha survives where the feature is supposed to be most valuable.
The signal may be more useful as a hold filter than a trading trigger
One of the strongest quantitative uses of supply-chain data may be to change when an existing position is retained rather than constantly creating new trades. A portfolio can keep a stock while its relationship score remains inside an acceptable region and only trade when the dependency changes enough to cross a wider threshold. This is conceptually similar to using a buy threshold that is stricter than the threshold required to continue holding an existing position. The strategy gives up some responsiveness in exchange for lower turnover and potentially higher net performance.
That design is especially sensible when the relationship data is slow moving. A quant can let fast alpha signals choose among securities while using supply-chain structure to prevent unnecessary exits from economically similar positions or to avoid adding another stock that repeats a major dependency. In that role, the data can improve the implementability of the portfolio without needing to predict next-period returns by itself. Gross alpha is only one way an alternative dataset can earn its place.
The conclusion is to optimize the net decision, not the prettiest backtest
A supply-chain strategy should be evaluated after the costs created by its actual portfolio rule. The researcher should compare rebalance frequencies, use realistic liquidity-sensitive assumptions, test hold bands, and separate slow structural features from fast event features. If the relationship signal only works when the portfolio trades faster than the information changes, the backtest is probably monetizing noise rather than economic structure. A slower implementation that preserves most of the information while cutting turnover can be much more valuable than a larger frictionless spread.
The walk-forward validation guide explains how to evaluate the strategy across genuinely unseen periods. The incremental-alpha guide explains how to test whether the network adds information beyond conventional predictors before execution costs are considered.
For relationship definitions and evidence limits, read the Altsets methodology.