Using observed Trackinsight ETF holdings and primary-market flows from June 2020 onward, the authors examine how cumulative ETF pressure relates to next-month stock returns. For U.S. thematic and sectoral ETFs they identify negative short-horizon and positive nine-to-twelve-month associations, then combine standardized twelve-month and three-month pressures into a stock-ranking signal evaluated in hypothetical historical and January–August 2026 post-sample portfolios. The dataset does not cover all global ETF assets, and the simple portfolio construction has substantial turnover. European equi

Money entering an investment fund reaches the stocks inside it. This paper asks a deceptively simple question: does when it arrives matter?

Same money. Different history.

An exchange-traded fund (ETF) holds a basket of investments. Creating or redeeming ETF shares changes demand for its underlying stocks. The researchers translate those flows into a stock-level measure of pressure.

The twelve-month total stays at 1.2 percent. Changing the recent share shifts pressure between the last three and the earlier nine months. Negative recent values represent redemptions.

Sorted by the recalculated signal. Higher scores are favored.

The paper’s signal compares a stock with the other stocks in the same month. A z-score says how far its pressure is above or below the cohort average, measured in standard deviations. The signal subtracts the three-month z-score from the twelve-month z-score.

This laboratory uses six fictional stocks, not market observations. The focus stock always has 1.20% total pressure; recent pressure is 1.20% × the slider share. The earlier nine months carry the remainder. Peers have fixed twelve-month pressures of 0.20, 0.45, 0.70, 0.90 and 1.50%, and three-month pressures of 0.20, 0.35, 0.20, 0.60 and 0.45%. Every change recalculates both means and population standard deviations across all six stocks. Bars spread each portion evenly within its window; that monthly shape is a teaching choice. Recent redemptions can coexist with a positive yearly total.

Keep the year’s total fixed. Which history gives this stock the higher signal?

Look backward. Read the next month.

The researchers measured how past ETF pressure relates to a stock’s return in the following month. Changing the history window reveals the two patterns behind the signal.

Each point is a separate regression estimate, controlling for the market’s return. A negative coefficient means higher past pressure was associated with a lower next-month return; a positive one means a higher return.

90% confidence interval: −1.71 to −0.48 · p < 0.01. The interval lies below zero.

For U.S. sector and theme funds, recent pressure is often associated with reversal. The 9–12 month estimates are all positive, with 90% intervals above zero: the longer history contains a different relationship.

Median ETF ownership of stocks in the study. Section 2.1.2 · Figure 1

What happens when the two clocks combine?

The paper ranks S&P 500 stocks using the difference between their twelve-month and three-month standardized pressure from sector and theme ETFs.

Each month, the exercise holds an equally weighted basket of the 50 highest-scoring stocks and shorts the 50 lowest-scoring stocks. The bars below show each stock basket’s return alongside the equally weighted S&P 500. The long–short strategy’s return is reported separately.

The high-score basket returned 13.44% annualized; the low-score basket returned −4.45%. The equally weighted S&P 500 returned 5.82%.

A different view of demand.

The combined flow signal captures who is receiving persistent ETF allocations relative to recent pressure. In the paper, it adds explanatory information beyond conventional factors and shows a low correlation with momentum, the tendency of past winners to keep winning.

Tables 7 and 9, pages 15–16. Adjusted R² measures how much of the variation the model explains, accounting for its number of variables.

Source

ETF Flows and Stock Market Dynamics: Patterns, Signals and Portfolio Construction Insights

Inès Barahhou and Sacha Abitbol · SSRN preprint · 2026 DOI: 10.2139/ssrn.7517178

The empirical study uses Trackinsight and Bloomberg data from June 2020, with monthly holdings and daily ETF flows aggregated monthly. Open and delisted ETFs are included; synthetic ETFs are excluded. Fund flows track primary-market creations and redemptions, rather than trading volume between investors. The paper’s authors performed the reported calculations.

The first laboratory is an original teaching model with a six-stock fictional cohort, the paper’s actual signal formula and a fixed twelve-month pressure. It shows a relative ranking, not an individual price forecast. The evidence chart reproduces numerical regression estimates and 90% confidence intervals from Tables 2, 3 and 10. The portfolio charts report the paper’s hypothetical portfolio exercise from Tables 4–6. January–August 2026 uses rules fixed in December 2025 and reports cumulative returns; the historical comparison reports annualized returns.

Source material: SSRN, all rights reserved. Paper figures are not reproduced. The interactive diagrams and charts are original compositions based on reported numerical facts. The graphical abstract is an AI-generated conceptual illustration informed by Figures 3 and 6, with an original composition. No additional supplement files were located.