Paper: arXiv 2403.17095
Authors: David Ardia, Clément Aymard, Tolga Cenesizoglu
Abstract
We reassess Boehmer et al. (2021, BJZZ)’s seminal work on the predictive power of retail order imbalance (ROI) for future stock returns. First, we replicate their 2010-2015 analysis in the more recent 2016-2021 period. We find that the ROI’s predictive power weakens significantly. Specifically, past ROI can no longer predict weekly returns on large-cap stocks, and the long-short strategy based on past ROI is no longer profitable. Second, we analyze the effect of using the alternative quote midpoint (QMP) method to identify and sign retail trades on their main conclusions. While the results based on the QMP method align with BJZZ’s findings in 2010-2015, the two methods provide different conclusions in 2016-2021. Our study shows that BJZZ’s original findings are sensitive to the sample period and the approach to identify ROIs.
Complexity vs Empirical Score
- Math Complexity: 4.5/10
- Empirical Rigor: 8.0/10
- Quadrant: Street Traders — empirical and applied, lower math complexity
Why this score: The paper focuses on empirical replication and robustness checks using real financial data, with moderate statistical analysis but no advanced mathematical derivations. It is highly data-driven with clear backtesting elements (e.g., long-short strategy profitability assessment), making it practical for implementation.
Research Flowchart
flowchart TD
A["Research Goal<br>Replicate & Extend Boehmer et al.<br>2021: ROI Predictive Power"] --> B
subgraph B ["Methodology & Data"]
direction LR
B1["Dataset 1: 2010-2015<br>(Original Period)"] --> B2["Dataset 2: 2016-2021<br>(Recent Period)"]
B1 --> B3["Method A: Original Method"]
B1 --> B4["Method B: QMP Method"]
end
B --> C{"Computational Analysis"}
C --> D["Regression Analysis<br>ROI vs. Future Returns"]
C --> E["Backtesting Strategy<br>Long-Short on ROI quintiles"]
D --> F{"Outcomes"}
E --> F
F --> G["2010-2015: Significant<br>Predictive Power"]
F --> H["2016-2021: Weakens Significantly<br>Large Caps: Insignificant<br>Profitability: Neutralized"]
Evidence Card
What the authors claim, on what data, and who has checked it. Read from the paper's own text by a language model: every quote is verbatim, nothing was re-run. How to read this card
Proposes
The paper reassesses Boehmer et al. (2021) on the predictive power of retail order imbalance and examines an alternative quote‑midpoint (QMP) method.
Source sentence
We reassess Boehmer et al. (2021, BJZZ)'s seminal work on the predictive power of retail order imbalance (ROI) for future stock returns.
Assumes 5
- Analyses focus on weekly horizons to reduce microstructure noise. statistical
Source sentence
focus on weekly horizons to reduce the impact of microstructure noise on [their] results
- Standard errors are computed using Newey‑West with six (or five) lags. statistical
Source sentence
Newey and West (1987) standard errors with six lags
- Trade frictions and transaction costs are ignored in the predictive analysis. execution
Source sentence
We ignore trade frictions and transaction costs here, and thus, the results do not have implications for whether outsiders can profit from these signals
- Order imbalances are measured based on share volume (Mroibvol). data
Source sentence
all discussions pertain to order imbalances based on share volume (Mroibvol for Mroib)
Show 1 more
- Retail order imbalance measures are used only as a signal, not to assess retail investors' own profits. other
Source sentence
Notice that this exercise uses marketable retail order imbalance measures merely as a signal to predict future stock returns, and thus, it provides no information on whether retail investors with marketable orders profit from their own trades
Data
- NYSE Trade and Quote (TAQ) data order-book commercial
Source sentence
introducing a novel algorithm for identifying retail trades within the NYSE Trade and Quote (TA…
- CRSP closing price returns market-prices commercial
Source sentence
Using order imbalances based on the number of trades (Mroibtrd) and/or CRSP closing price returns does not fundamentally change the interpretation
Costs
Not modeled
Source sentence
We ignore trade frictions and transaction costs here, and thus, the results do not have implications for whether outsiders can profit from these signals
Finds 5
- The predictive power of ROI weakens significantly in 2016‑2021; past ROI no longer predicts weekly returns on large‑cap stocks and the long‑short strategy is no longer profitable. negative
Source sentence
We find that the ROI's predictive power weakens significantly. Specifically, past ROI can no longer predict weekly returns on large-cap stocks, and the long-short strategy based on past ROI is no longer profitable.
- The coefficient and t‑stat of the first lag (Mroibvol) drop about 50% and 20% respectively in the recent period. statistical
Source sentence
the coefficient estimate and the t-stat of the first lag, Mroibvol(i, w −1), are 50% (0.0983 vs. 0.1982) and 20% (57.30 vs. 71.81) lower in 2016-2021 period
- Predictive regression coefficient for future returns falls 34% and its t‑stat 48% in 2016‑2021. statistical
Source sentence
the coefficient estimate of Mroibvol(i, w −1) and its t-stat are 34% (0.0006 vs. 0.0009) and 48% (7.95 vs. 15.14) lower
- Alphas for one‑ and two‑week horizons are more than 65% lower in the recent period; many become statistically insignificant. performance
Source sentence
for one- and two-week horizons, 2016-2021 alphas are more than 65% lower (0.143% vs. 0.437% and 0.177% vs. 0.613%)
- The QMP method yields similar conclusions for 2010‑2015 but shows stronger predictability for most subgroups in 2016‑2021 compared to the original method. other
Source sentence
both lead to similar conclusions and economic magnitudes in both periods (Panel (b) vs. (a) and Panel (d) vs. (c))—with some notable differences in the recent period such as significantly stronger one-week small…
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 22 related papers). Coverage is partial, so this is neither replication nor refutation.
Code
No repository by the authors found in the paper.
Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.