Earnings call participation network and dynamics as an alternative data factor for US equities.

01Participation as a source of information

An earnings call records what a company reports and which external analysts ask questions. This study asks whether the history of that participation can improve stock selection beyond a model built from prices and trading volumes.

External analysts participate in a professional research capacity. Their work brings sector knowledge, familiarity with company reporting and comparisons across businesses into the discussion. A question may reflect extensive analysis, expectations and unresolved concerns formed before the call. Analysts differ in experience, expertise and access, so the information associated with their participation need not be equivalent.

When the same analysts recur across companies and reporting periods, their participation forms an observable network. Its changing connections provide a way to study how professional attention is distributed and how its focus shifts. ECPND converts this history into a daily research factor. Our hypothesis is that participation patterns contain information relevant to subsequent returns that is not fully captured by a model built from prices and trading volumes.

What the research suggests

Earlier work establishes two relevant connections. Mayew, Sharp and Venkatachalam (2013) find that analysts asking questions issue more accurate and timely annual earnings forecasts immediately after calls than nonparticipants [1]. Mayew (2008) finds that analysts with more favourable stock recommendations are more likely to obtain access to questioning [2]. Participation can therefore reflect both research activity and the opportunities management provides to engage.

Research published in the 2020s extends this picture. Bradley, Gokkaya and Liu (2020) associate professional connections between analysts and executives with better forecast accuracy, more informative recommendations and greater participation in calls [3]. Rennekamp, Sethuraman and Steenhoven (2022) examine the interactions themselves: greater conversational engagement, measured through linguistic style matching, is associated with larger absolute price movements during the discussion [4]. This supports studying interaction quality alongside participant identity.

Two studies published in 2026 address the significance of analyst expertise more directly. Awyong, Cho and Yang find that analysts' share of the discussion is associated with subsequent revisions to management forecasts, with stronger associations when the analysts are more experienced and have greater forecasting ability [5]. Holowczak, Jiu, Kyung and Yu link all-star analyst participation to abnormal returns during the Q&A session, with a stronger relationship for analysts whose forecasts are more accurate [6]. These findings give an empirical basis for considering who participates, as well as how many people participate.

The studies examine different measures and outcomes from ECPND. They motivate its economic rationale; the score’s contribution to future portfolio returns requires its own test. Company size, sector conditions, coverage practices and access to management may also shape the observed network. Our paired experiments hold the underlying ranking, capital, execution rules and costs constant, then measure what changes when ECPND is used in stock selection. Separating the economic mechanisms behind any contribution remains a further research question.

03From participation to ECPND

The score summarizes eligible participation history as a daily measure, indexed by date and ticker for use alongside stock rankings in systematic equity research.

From source observation to portfolio comparison

1. Observe. Read the issuer, call date and named external participants from structured transcript components.
2. Connect. Normalize recurring identities and connect eligible participation observations across companies.
3. Score. Calculate a bounded value from eligible participation history, preserving missing observations.
4. Evaluate. Apply the stated score-priority rule and compare the resulting portfolio with its matched baseline.

Participation refers to named external analysts identified in the transcript’s structured speaker records. Management and issuer representatives are excluded. Recurring identities link participation across companies and over time. The score evaluates this history using outcomes from other companies, with return periods completed before the scoring date. The issuer’s own outcomes are excluded from these comparisons, and estimates based on limited histories receive less influence.

• Field: ecpnd_score, a daily value between 0 and 1.
• Direction: Higher values are more favourable; neither a probability nor a return forecast.
• Persistence: Daily scores combine the latest eligible call with evolving network evidence.
• Missing score: No eligible value for that stock and date: blank in CSV, null in JSON.

Daily scores and network evolution

ECPND evaluates each company through its connections to the wider earnings-call participation network. The company’s latest eligible call identifies the participants relevant to its score, while their eligible history across other companies provides the broader context. New calls refresh participation records, and completed outcomes from earlier calls update the evidence associated with shared participants. These changes can affect several connected companies, allowing scores to evolve even when an individual company has no new call.

The freshness limit applies to the company’s own latest eligible call record, which expires at an age of 126 US trading sessions, roughly six months. Age starts at zero on the first US trading session after the call date; recalculation does not reset it. A newer eligible call replaces the earlier record. If the new record lacks usable participants or sufficient history, the score is missing; the calculation does not fall back to an older call. An expired record likewise cannot support a current score.

In the historical score file, date identifies the session for which the value was calculated under the study’s timing convention. Use the supplied score for each portfolio decision session. Buy & hold retains its initial positions despite later score changes or expiry. Weekly rebalancing uses the scores for each scheduled decision and applies the stated missing-score policy.

Live snapshots may incorporate calls received during the current day, while historical scores apply next-session eligibility. For live decisions, follow the publication-time checks and snapshot-retention workflow in Section 2.

What does “missing score” mean?

A missing score describes the availability of a measurement. It means the inputs eligible under the declared timing convention do not support an ECPND value for that stock on that date. It is not a negative opinion about the company, a neutral score, or a forecast of poor returns. A numerical score of 0 is a valid value and is different from a blank.

• No eligible prior call: No source call is eligible for that stock and date.
• No usable participants: The latest eligible call has no identifiable external analysts.
• Insufficient history: Eligible peer history does not meet the evidence requirements.
• Outdated call record: The latest record has reached the 126-session expiry limit.

A day without a new company call is therefore not automatically a missing-score day. Use the supplied value for the decision date and preserve missing observations. Substituting an earlier score or zero would change the stated selection rule.

Treatment in the portfolio tests. Stocks without an ECPND score remain eligible under the baseline ranking. They receive no factor penalty and do not enter score-percentile calculations. This treatment is identical across the three main baselines, both policies and every threshold. A separately labelled scored-only control, documented with the unfiltered study, removes missing-score stocks from both arms before ranking.

Scope of disclosure. This paper documents the factor’s economic rationale, data inputs, timing conventions, study design and results. The detailed methods for resolving participant identities and converting participation history into ECPND scores remain proprietary. The common application rule in Section 6 specifies how supplied scores affect every tested threshold, including percentile ranking and missing-value treatment. Researchers can use these rules to evaluate the delivered factor without reproducing its underlying calculation.

04Three baselines, one portfolio rule

ECPND is tested as an overlay. Each baseline ranks the same stocks on its own; the paired portfolio uses the same ranking and lets ECPND change selection priority. Dates, capital, execution and costs are identical within each pair.

Custom Qlib model

Microsoft's Qlib framework [7] supplies 157 price and volume features from Alpha158 [9]. LightGBM [8] learns from them to rank stocks by their expected return over the following 42 sessions. The model is refitted once a year on an expanding history, using only outcomes completed before the test year, with four fixed training seeds. It receives no ECPND input.

Qlib reference model

The second baseline uses Qlib's published Alpha158 and LightGBM configuration with its original next-day target and published hyperparameters, adapted to US stocks. It is a model specification we did not design ourselves. It is also fitted once a year with the same four seeds and receives no ECPND input.

Momentum

The third baseline needs no model. Stocks are ranked by their return from 252 to 21 sessions before the decision, the standard twelve-month momentum that skips the latest month [13]. It gives one deterministic ranking per date and asks whether participation information adds value beyond a simple price rule.

One portfolio rule for all three

• Portfolio: 50 equally weighted long positions across 95% of USD 1 million.
• Decisions: weekly, executed at the opening price, using rankings from the preceding close.
• Costs: 10 bp on every executed buy and sell.
• Policies: buy & hold keeps the initial positions; weekly rebalancing repeats the selection each week.
• Horizons: 3, 6 and 12 calendar months, measured for a new portfolio pair started in each of the 245 weeks.

Results are paired differences: the return of the portfolio with ECPND minus the return of its baseline, in basis points. For the two Qlib models the four seeds are averaged within each starting week. Each 95% interval comes from 5,000 circular-block bootstrap resamples of the starting weeks.

05What we investigated, in order

The application in this paper is the end of a sequence of investigations on the same history. This section lists every step in the order it was carried out, with its main result, including the results that were weak or unfavourable. The full tables are documented in separate research notes.

First study: one model, one rule

The first study added ECPND to the custom Qlib model only, giving lower priority to the lowest 20% of scores, for weekly portfolios started from January 2022. Six-month buy & hold portfolios gained +54 bp, with a 95% interval of +13 to +99 bp. The other five comparisons were positive, but their intervals included zero.

The score on its own

We then tested the score without any model. The rank correlation between score and later return was positive at every horizon, +0.015 over 63 sessions and +0.040 over 126 sessions, but none was statistically significant. A portfolio selected by score alone grew by 15.4% a year after costs, compared with 7.1% for the custom model and 20.6% for momentum on the same stocks.

Known factors and sectors

Regressions of the return differences on the standard equity factors (market, size, value, profitability, investment and momentum) produced no significant alpha in any of 16 tests; the smallest p-value was 0.10. After controlling for sector and price characteristics, the relationship between score and return remained positive and remained insignificant.

More baselines

Across all four model seeds, the 20% rule raised the growth of the custom Qlib accounts by 1.29 percentage points a year on average, while a momentum account lost 0.25 points. A second, published Qlib model with daily trading from 2023 did not benefit: its accounts lost 0.29 points a year on average, and momentum under the same daily rules lost 0.54 points. At this stage ECPND helped one model and not the others.

How strongly should the factor act?

We tested every threshold from 0% to 100% in five-point steps. On the daily accounts, stronger settings did better: at 60%, momentum rose from 26.8% to 28.7% a year and the published Qlib model from 10.6% to 13.2%. Very high thresholds, however, increasingly select stocks that have no score, so they say less about the score itself. A review of rolling one-year windows found no setting that improved every period, every seed and every drawdown.

Three baselines under one protocol

To compare like with like, all three baselines were rerun on identical stocks, dates and weekly rules from January 2022, again over the full threshold grid. At 20%, six-month buy & hold portfolios gained +53 bp, +20 bp and +41 bp over the custom, reference and momentum baselines, and 15 of 18 mean differences were positive. The effect was small: one comparison remained significant after a Holm correction, and none after a correction across the whole threshold grid.

Does the result depend on weekly trading?

Continuous accounts that rebalance every week gave a mixed picture on the unfiltered pool: +730 bp of total return for the custom model, -357 bp for the reference model and -33 bp for momentum over the full period. We then rebalanced quarterly, half-yearly and yearly, each with staggered starting weeks. All nine mean differences were positive and eight exceeded their weekly control, but every setting also had at least one starting week with a negative contribution.

Is it momentum of connected companies?

Companies that share analysts tend to move together, so the score could simply repeat the momentum of connected companies. The score's rank correlation with that linked momentum is +0.49. After controlling for it, the relationship between score and return stayed positive but was not significant. Added to a portfolio built on linked momentum itself, the 20% rule raised growth from 6.3% to 8.8% a year, also without statistical significance. The two sources of information overlap, and the tests could not separate them.

A broader universe and a longer history

On roughly 1,900 liquid US stocks per week instead of about 460, the relationship between score and return over 126 sessions was the same as in the original universe (+0.045 in both). Going beyond the S&P 500-based universe neither strengthened nor weakened it. Extending the history back to March 2020 was unfavourable: in 2020 and 2021 the relationship was negative (-0.12 over 126 sessions), and over the whole extended period it was close to zero.

Costs and missing scores

Further controls reran the common six-month comparison with trading costs of 0, 10, 25 and 50 bp. Buy & hold differences did not change with costs, because both portfolios trade once. With weekly rebalancing, higher costs did not reduce the contribution; the one negative case, the reference model at -22 bp, turned positive only at 50 bp. Restricting both portfolios to stocks that have a score left all six differences positive, between +2 and +60 bp.

Where is the contribution concentrated?

Finally we asked whether the contribution differs with a stock's recent volatility. Volatility was measured over 21, 63 and 126 sessions; candidate pools kept the most volatile 50%, 40%, 30% or 20%, and five separate quintiles were compared, 20 settings in total. The pattern was clear in one direction: in all 18 comparisons the factor contributed more in the most volatile fifth than in the least volatile fifth. The pools in between were less regular: with buy & hold, the contribution of both Qlib models rose step by step as the pool was narrowed, while momentum did not follow that pattern (Section 9).

The resulting application

On the full universe, ECPND was a small and uneven contribution that could not be separated from related signals. Among volatile stocks it was larger and more consistent. The application presented here therefore combines two choices: the 20% rule, applied among the most volatile fifth of stocks over 63 sessions. Both choices were made after seeing historical results. The following sections show what this application delivered, and Section 10 states the limits that follow from choosing it in hindsight.

Two lines of work continue outside this paper: the combination of ECPND with the text factor Earnings Call Quantitative Dynamics and Intensity (ECQDI), and a forward evaluation that records delivered scores and portfolio decisions as they happen.

06The 20% rule inside the volatility filter

The application has two steps at every weekly decision. The volatility filter defines the pool; ECPND then changes priority inside that pool. Baseline and factor portfolio always use the same filtered pool.

1. Measure volatility. For each eligible stock, take the 63 daily returns ending at the preceding close and annualize their standard deviation. Nothing from the decision day or later is used.
2. Keep the most volatile fifth. Rank all eligible stocks by this value and keep those above the 80th percentile. The pool has 91 to 100 stocks per week, out of 453 to 497 eligible stocks.
3. Rank ECPND inside the pool. Stocks whose score is in the lowest 20% of the pool receive lower priority. Stocks without a score, on average 3 per week, receive no penalty.
4. Select 50 stocks. Order the stocks without a penalty by the baseline ranking, followed by the penalized stocks in the same order, and take the first 50. The baseline portfolio takes the first 50 of the pool without any penalty.

The factor does not pick stocks on its own. It removes weakly scored candidates from the front of a ranking the baseline has already made. Because the pool holds roughly twice as many stocks as the portfolio, the 20% rule typically replaces a handful of the 50 positions.

The annualized volatility needed to enter the pool varied between 29% and 52% over the study, with a median of 36%. A volatility rank is only meaningful within its reference universe: here, the eligible large-cap stocks on that date.

07Portfolio results

Across three baselines, two policies and three horizons, all 18 mean return differences are positive. In 14 of them the 95% interval lies entirely above zero.

Figure 1Average contribution across the three baselines

On average over the whole tested period and across the three baselines, ECPND added +37 bp, +80 bp and +172 bp with buy & hold and +60 bp, +130 bp and +280 bp with weekly rebalancing over 3, 6 and 12 months. The uncertainty ranges of all six results lie entirely in positive territory. The contribution grows with the holding period and is larger with weekly rebalancing at every horizon. Over twelve months with weekly rebalancing, even the lower end of the range is above +200 bp.

Mean return difference in basis points with its 95% interval. In 14 of the 18 individual comparisons and in all six averages the interval lies above zero. The four intervals that include zero are Custom Qlib with buy & hold over 3 months and momentum with buy & hold over all three horizons. Completed starting weeks: 232, 219 and 193.

With weekly rebalancing, every interval is above zero. Six-month portfolios gained +119 bp, +173 bp and +98 bp; twelve-month portfolios gained +227 bp, +363 bp and +250 bp. The contribution grows with the horizon in every row.

Buy & hold results are positive but less certain. Both Qlib models show clear six- and twelve-month contributions. For momentum, the three buy & hold intervals include zero, and so does the three-month interval of the custom model. These four comparisons are the weakest part of the evidence.

Testing 18 comparisons at once raises the chance that some look significant by accident. After a Holm correction across all 18, 5 remain significant at the 5% level: the six- and twelve-month results of both Qlib models and the twelve-month result of momentum, all with weekly rebalancing. No buy & hold comparison passes this stricter test.

Every difference is measured against a baseline that draws from the same filtered pool. It isolates what changes when ECPND is added, not the return from choosing volatile stocks.

08Continuous accounts

The cohort results average many overlapping portfolios. A continuous account is the simpler view: one account per baseline, funded once in January 2022 and run until September 2026, rebalanced every week, every month or every quarter.

Qlib rows are means of four separately funded seed accounts; momentum is one account pair. All accounts start on 10 January 2022 and include all fees.

ECPND raised the annual growth rate in all nine accounts, by between 1.1 and 3.6 percentage points. For both Qlib models, all four seeds improved at every rebalancing frequency. The contribution does not depend on weekly trading: it is of similar size with monthly and quarterly rebalancing. Maximum drawdown was smaller or practically unchanged with the factor, so the higher return did not come with deeper losses in this period.

These are single historical paths over one market period, all started on 10 January 2022; other starting dates may give different results. They include all fees and are not averages of the cohorts in Section 7.

09Stability across years, volatility and settings

A result that rests on one good year, one threshold or one cost assumption would say little about the factor. This section looks at how the contribution is spread across starting years, how it changes with the volatility of the stocks, and how it responds to other thresholds and trading costs.

Across starting years

Figure 2Contribution by starting year

Figure 2 groups the portfolios by the year in which they were started. Of 84 year groups across all baselines, policies and horizons, 70 are positive, and in every starting year the positive groups are the clear majority. The contribution is therefore not carried by a single year. It is uneven, though: 8 of the 14 negative groups belong to momentum, including its weekly rebalanced portfolios started in 2025 and 2026. Both Qlib models have 3 negative groups each, none larger than -30 bp.

Scales differ between horizons. Year groups that have not yet completed a horizon are left out, and the 2026 groups contain only 24 starting weeks at three months and 11 at six months.

Across volatility

Figure 3Factor contribution across volatility pools

Figure 3 applies the same 20% rule to pools of different volatility. In the least volatile fifth the six-month contribution is close to zero or negative for buy & hold. In the most volatile fifth it is positive for every baseline and policy. For both Qlib models with buy & hold it rises with each narrower pool; momentum and the weekly policies are less regular in between, and some intermediate pools are negative.

Difference between the ECPND contribution in the two pools, in basis points, with its Holm-adjusted p-value across the 18 comparisons. All 18 differences are positive; 4 have a Holm p-value below 0.05, all at twelve months.

Table 4 compares the two ends week by week. The contribution is larger in the most volatile fifth in all 18 comparisons, and 4 of them remain significant after a Holm correction, all at twelve months. This compares two different stock pools: it shows where the factor contributed, not that volatility causes the contribution.

Across thresholds and trading costs

Counts refer to the 18 comparisons across three baselines, two policies and three horizons. Six-month values are three-baseline averages in basis points; the last column averages all 18 mean differences.

Table 5 repeats the application with other thresholds inside the same filter. Every threshold from 10% to 50% produced positive mean differences in at least 17 of 18 comparisons. The average contribution grew with the threshold up to about 50% and fell beyond it; above 90% it turned negative. The 20% rule is therefore not an isolated peak. It has the largest number of intervals above zero, while stronger thresholds delivered larger but less certain contributions.

With trading costs of 0, 10, 25 and 50 bp per executed trade, all 18 mean differences stayed positive at every cost level, and between 13 and 14 intervals stayed above zero. Buy & hold differences do not depend on the cost level, because both portfolios trade once. With weekly rebalancing, the six-month contribution at 50 bp was +212 bp, +320 bp and +92 bp for the custom, reference and momentum baselines.

10What the result does not show

• Chosen in hindsight. Both the 20% threshold and the volatility filter were selected after reviewing results on this same history. The intervals in Section 7 describe each comparison on its own and are not corrected for that search. The result is a documented historical application, not a pre-registered test.
• One market period. The study covers January 2022 to September 2026. The 245 weekly starts overlap heavily and share one market history; they are not independent experiments.
• A narrow, volatile pool. The application selects 50 of roughly 96 highly volatile large-cap stocks. Capacity, spreads and market impact in such a portfolio are not modelled beyond the flat 10 bp cost.
• Not every comparison is clear. Momentum with buy & hold and the three-month custom buy & hold comparison have intervals that include zero. After a Holm correction, 5 of 18 comparisons remain significant, none of them buy & hold.
• No claim of independent alpha. On the full universe, the score's relationship with returns did not remain after controlling for the momentum of analyst-linked companies. The same control has not been run inside the volatility filter.
• Not the strongest threshold. Thresholds around 45% to 50% had larger average contributions than 20% in this history. Choosing among thresholds after the fact adds to the search described above.

Evidence on new data comes from the forward evaluation, which records delivered scores and portfolio decisions as they happen.

11Research records and further papers

A separate internal recalculation reproduced all volatility values, cohort summaries and continuous accounts of the volatility study and rebuilt the weekly selections independently. The largest difference in account value was below one hundredth of a cent. This verifies the arithmetic; it is not an external audit of the price data.

The research package is deliberately small. It contains two years of daily scores for both factors, released with a six-month delay, one worked selection example that shows each stock's volatility next to both scores, and a README. These scores are shorter than the study history and do not reproduce it in full. The calculation of the score itself remains proprietary. The detailed study records are kept in our research archive.

The detailed tables for the threshold grid, rebalancing frequencies, score diagnostics and controls are documented in separate research notes. The combination of ECPND with ECQDI is the subject of its own paper.