{"id":"ff29b747-9388-44e3-91fb-208fc5c12aee","arxiv_id":"2506.00206","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":6,"one_line_summary":"High value-to-price stocks earn larger one-to-three-year returns than low V/P stocks, and Fama-French five-factor alphas remain positive for the highest V/P quintile in years two and three.","lead":"This paper tests whether portfolios sorted by a residual-income-based value-to-price (V/P) ratio earn excess returns over one to three years in US stocks. It reports that high V/P portfolios beat low V/P portfolios on raw and size-adjusted buy-and-hold returns, and that standard factor models leave positive alphas for the highest V/P stocks in years two and three.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Long-horizon V/P factor regressions rely on OLS t-statistics that ignore the 12-month block structure of annually rebalanced portfolios; the Q5 five-factor alphas may not survive Newey-West or cohort-clustered inference.","rationale":"The paper is a careful empirical extension of the V/P anomaly, with a detailed valuation model, out-of-sample jackknife estimation, and multi-horizon portfolio analysis. The reader's conditional verdict is appropriate. The load-bearing weakness is inference on the long-horizon alphas: the monthly return series used in Tables 5-8 are not draws from a stationary i.i.d. process because the underlying portfolios are held in fixed 12-month blocks. The paper itself acknowledges the overlap problem when computing buy-and-hold return significance in Table 1, but does not carry that correction into the factor-model tables. Since the marginal significance of the Q5 five-factor alphas (t close to 2.3-2.7) is exactly what separates the paper's claim from the prior literature, an inference correction could flip the main result. Other concerns, such as equal-weighted microcap tilts or multiple testing, are real but secondary: the paper reports GRS joint tests and the raw spreads are large, so the alpha significance is the decisive issue. The proposed test directly targets this vulnerability without requiring re-estimation of the valuation model.","tokens_in":41891,"tokens_out":10500,"duration_ms":112575,"concrete_test":"Re-estimate the five-factor regressions for the Q5 portfolio in Tables 5-6 (and the corresponding GRS tests) using Newey-West standard errors with 12 and 24 lags, and also with standard errors clustered by formation-year cohort (23 clusters). Check residual autocorrelation at lags 1-12. If the year-2 and year-3 Q5 alphas remain significant at the 5% level under these corrections, the central claim survives; if not, the reported significance is an artifact of overstated OLS precision.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim that Fama-French five-factor model cannot explain all V/P excess returns, especially in years 2 and 3, rests on alphas such as Q5 FF5 alpha of 0.004-0.005 per month with t = 2.34-2.66 (Tables 5-6). These t-statistics are computed from monthly time-series regressions that treat each month as an independent observation. But the monthly return series for overlapping, second-year, and third-year portfolios are built from formation cohorts that are rebalanced only once per year: each 12-month block is the return path of the same set of stocks. Within a block, residuals are likely autocorrelated, and the effective number of independent formation cohorts is roughly 23, not 276. The paper applies Newey-West corrections in Table 1 for overlapping buy-and-hold returns, explicitly noting the overlap problem, yet Tables 5-8 report only OLS inference. A modest positive autocorrelation in residuals would be enough to push the reported t-statistics below conventional significance levels. If those alphas are not robust to serial-correlation-consistent or cohort-clustered standard errors, the conclusion that standard factor models cannot explain long-horizon V/P returns loses its empirical support.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper constructs a residual-income value-to-price (V/P) measure using the Feltham-Ohlson/Ohlson linear information dynamics estimated by out-of-sample industry-year SUR, forms US quintile portfolios, and tests whether V/P-sorted returns are explained by the CAPM, the Fama-French three-factor model, and the Fama-French five-factor model over one-, two-, and three-year horizons. The main empirical findings are: (i) high-V/P portfolios outperform low-V/P portfolios, with spreads increasing from 12 to 36 months; (ii) at the firm level, V/P retains a positive coefficient after controlling for risk proxies, though only marginally in the full model; and (iii) five-factor alphas for the highest V/P portfolio are significant in overlapping returns and in years 2 and 3, leading the authors to conclude that standard factor models cannot fully explain the V/P effect. The paper also studies dividend-adjusted returns and the characteristics of high-V/P firms.","tokens_in":42210,"tokens_out":9691,"duration_ms":98350,"significance":"If the long-horizon results hold, the paper makes a useful contribution by extending the V/P anomaly literature beyond the usual first-year returns and by showing that profitability and investment factors do not absorb high-V/P returns in years 2 and 3. The V/P measure is constructed from accounting data and analyst forecasts via an out-of-sample SUR procedure that does not use the returns used for testing, so circularity is not a concern; the factor models are external benchmarks. The paper's explicit use of Newey-West corrections in Table 1 for overlapping buy-and-hold returns is a positive feature, and the literature review is extensive. However, the central long-horizon claim depends on alphas whose reported t-statistics come from OLS regressions on overlapping annual cohorts, and this inference issue must be resolved before the contribution can be assessed.","major_comments":[{"comment":"The Q5 five-factor alphas that carry the paper's central claim—overlapping returns of 0.004 (t=2.34) in Table 5 Panel C, and second- and third-year returns of 0.004 (t=2.37) and 0.005 (t=2.66) in Table 6 Panel C—are based on monthly OLS regressions of annually rebalanced portfolio return series. Because each annual cohort is held for 12 months or longer, the residuals have a 12-month block structure, and the effective number of independent formation cohorts is about 23 rather than 276. The paper applies Newey-West corrections for overlapping buy-and-hold returns in Table 1 but not to these factor regressions. Please report standard errors clustered by formation year (or Newey-West with at least 12 monthly lags) for all alphas and factor loadings in Tables 5–8, and adjust the GRS F-statistics accordingly. If the Q5 alphas in years 2 and 3 are no longer significant after this correction, the conclusion that 'these factors cannot explain all variation in excess returns especially for years two and three' loses its empirical support.","section":"§5.3.1–5.3.3, Tables 5–8"},{"comment":"In the full model of Eq. (4), the V/P coefficient is 0.013 with t=1.9 and is displayed with two stars, but with 11,693 observations the two-sided p-value is approximately 0.057, so the coefficient is not significant at the 5% level. The text's statement that 'the coefficient on V/P remains significant and positive' and that 'omission of risk factors is not a likely explanation of the V/P effect' is therefore overstated. Additionally, Ret36 for adjacent firm-years overlap by up to 24 months, so the OLS standard errors in Table 4 may understate uncertainty even beyond the marginal t-statistic. Please report exact p-values, use one-sided tests only with explicit justification, and consider clustering by firm or formation year.","section":"§5.2, Table 4, Eq. (4)"}],"minor_comments":[{"comment":"The reference 'Frankel and Lee (1988)' appears to be a typo for 'Frankel and Lee (1998)' and should be corrected.","section":"§2.3"},{"comment":"The definition of overlapping returns in the footnotes of Tables 7 and 8 says the portfolio consists of first-year returns from year t, second-year returns from year t-2, and third-year returns from year t-3, but Tables 5 and 6 describe the cohorts as t, t-1, and t-2; this inconsistency should be fixed.","section":"Tables 7–8 footnotes"},{"comment":"The table title says the period is 1987–2000, but the rows extend to 2014; the title should be updated to reflect the full sample period.","section":"Table A1"},{"comment":"The row 'Years' reports 1993–2014 for Model I and 1992–2014 for Model II; the discrepancy should be explained or aligned.","section":"Table 3"},{"comment":"The text refers to 'Fame and French (2015)' in one instance; this is a typo for 'Fama and French (2015)'.","section":"§5.3.1"},{"comment":"The Altman Z-score formula in the Table 2 notes uses coefficients 0.012, 0.014, 0.033, 0.006, and 0.999, while the appendix uses the standard 1.2, 1.4, 3.3, 0.6, and 1.0; the table note should be corrected for consistency.","section":"Table 2 notes"}],"recommendation":"major_revision","confidential_remarks":"The paper addresses an important question and the V/P construction is careful. The main obstacle is statistical inference for the long-horizon alphas; if the authors can show that the Q5 alphas survive cluster-robust or Newey-West corrections, I would be supportive. I would also encourage them to include the discount-rate robustness checks in the main tables rather than only in the text."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Two things to know. First, this paper delivers a clean replication of the V/P anomaly: portfolio spreads over one, two, and three years are large, monotonic, and survive size adjustment, and the authors are careful to apply Newey–West corrections in the buy-and-hold comparisons. Second, the genuinely new part—Feltham-Ohlson SUR valuation with conservatism, year-two/year-three factor analysis, and dividend-adjusted returns—is worth taking seriously, but the claim that the five-factor model cannot explain years two and three depends on OLS t-statistics for overlapping monthly portfolios, and that inference is not yet credible.\n\nWhat the paper does well: the valuation procedure is out-of-sample (jackknife), so V/P is not fit to the returns it predicts. The replication of Frankel and Lee is done with a longer sample, and the comparison to B/M is useful. The GRS tests and factor-loading patterns are informative. The paper makes an honest attempt to distinguish mispricing from risk using firm-level regressions, and the literature coverage is thorough.\n\nWhere it is soft: the main hole is the one the stress-test note identifies. The overlapping portfolios in Tables 5–8 are formed from cohorts rebalanced annually, so each month's return shares stocks with the previous year's portfolio. That mechanically induces serial correlation at lags up to 23 months. The paper applies Newey–West in Table 1 but not in the factor regressions. The t-stats on Q5 alphas around 2.3–2.7 would likely shrink under HAC or cohort-clustered errors; whether they remain significant is an open question. The firm-level V/P coefficient in the full model has t=1.9, which is marginal. These are fixable with a few lines of code, but they need to be fixed before the “cannot explain all variation” conclusion can stand. Minor issues include some typos (“weekly” for “weakly,” “corelated”) and a repetitive literature review.\n\nBottom line: this paper is a solid empirical exercise that mostly replicates known results. The incremental contribution—years two and three and dividend-adjusted alphas—is plausible but not yet established because of the inference problem. It deserves a serious referee; I would send it out with a requirement that the authors re-run the factor regressions with serial-correlation-robust or cluster-robust standard errors and report whether the Q5 alphas survive. If they do, this is a nice paper. If they don't, it's a replication with a weaker extension, but still cite-worthy as a careful out-of-sample test.","headline":"Replicates the V/P anomaly carefully, but the new years-2/3 factor alphas rely on overlapping-return OLS that needs HAC correction before the central claim is trustworthy.","tokens_in":42712,"tokens_out":2533,"would_cite":true,"duration_ms":26086,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper claims that portfolio sorts on the residual-income value-to-price (V/P) ratio generate excess returns over one to three years, and that the Fama–French five-factor model cannot explain all of these returns, especially for…","keywords":["residual income valuation","value-to-price ratio","V/P anomaly","value investing","asset pricing factors","mispricing","long-horizon returns","Fama-French five-factor model"],"falsifier":"Re-estimate the five-factor alphas for the highest-V/P portfolio in years two and three using Newey–West or two-way clustered standard errors on the overlapping monthly return series; if the t-statistics fall below conventional thresholds, the paper's claim that factors cannot explain all variation loses its statistical support.","tokens_in":41713,"feed_emoji":"📈","tokens_out":4459,"duration_ms":50426,"temperature":0.7,"pith_summary":"The paper tests a simple accounting-based value signal: the ratio of residual-income-model intrinsic value to market price (V/P). It finds that high-V/P portfolios beat low-V/P portfolios by 5.3%, 13.7%, and 27.8% in raw buy-and-hold returns over 12, 24, and 36 months. After adjusting for the Fama–French five factors, the top V/P quintile still earns monthly alphas of about 0.004 to 0.005 in overlapping, second-year, and third-year returns. The paper argues this persistent unexplained return favors a mispricing interpretation over a pure risk explanation, and that standard factor models leave a meaningful part of the value-to-price anomaly unexplained.","feed_headline":"Value-to-price stocks beat risk models into year three","feed_subtitle":"A simple accounting value signal keeps earning risk-adjusted returns in years two and three, US data show.","key_machinery":"The key machinery is the residual income valuation model of Ohlson (1995) and Feltham and Ohlson (1995), implemented as a system of forecasting and valuation equations (book value, abnormal earnings, other information, and market value) estimated jointly by seemingly unrelated regression. Intrinsic value V is computed out-of-sample for each firm-year-industry using a jackknife procedure, and V/P is then used to form quintile portfolios. The paper's long-horizon evidence comes from an overlapping portfolio construction that holds current, prior-year, and two-year-old cohorts in each month, and from regressions of these portfolio excess returns on the CAPM, Fama–French three-factor, and five-factor models, with the Gibbons–Ross–Shanken statistic used to test whether all alphas are jointly zero.","core_discovery":"The central discovery is that the V/P effect is not a one-year phenomenon: high-V/P stocks keep earning risk-adjusted excess returns in the second and third years after portfolio formation. In the paper's numbers, the highest-V/P quintile produces a five-factor alpha of roughly 0.004 to 0.005 per month (t-values about 2.34 to 2.66) in overlapping, year-two, and year-three returns, and these alphas rise when dividends are included. The paper also shows that V/P predicts 36-month buy-and-hold returns at the firm level after controlling for beta, size, idiosyncratic volatility, leverage, distress, profitability volatility, and book-to-market. Because the highest-V/P stocks load positively on the investment factor in year two and on the profitability factor in year three yet still earn positive alpha, the paper concludes these stocks are significantly mispriced relative to their future equity growth and profitability persistence.","pith_inferences":["The reported t-statistics for year-two and year-three alphas come from overlapping portfolios, which mechanically induces serial correlation in monthly returns; correcting standard errors for this overlap could weaken the significance of the long-horizon alphas.","The paper's own characteristics of high-V/P stocks (small size, high idiosyncratic volatility, low analyst coverage) align with limits-to-arbitrage explanations, suggesting the mispricing story could be tested further by conditioning on arbitrage frictions.","A natural extension is to test whether the V/P alpha survives the q-factor model or a model that includes a separate V/P mimicking factor, and whether similar long-horizon alphas appear in international markets.","Because high-V/P stocks load on investment and profitability factors in ways that vary by year, decomposing the anomaly around earnings announcements could reveal whether the post-year-one returns reflect slow fundamental information diffusion."],"forward_implications":["If the V/P anomaly persists into years two and three, a simple accounting-based value signal can be used to build multi-year value strategies that outperform book-to-market sorting.","The five-factor model's failure to price high-V/P stocks in later years implies either an omitted risk dimension or a slow correction of mispricing, both of which invite further asset-pricing work.","Profitability and investment factors matter for middle-V/P portfolios but not for the extreme high-V/P portfolio, suggesting the high-V/P premium is not just the standard profitability or investment premium.","Firm-level regressions showing V/P predicts 36-month returns after controlling for common risk proxies strengthen the case that the V/P effect is a distinct anomaly, not a restatement of size or value effects.","Dividend-adjusted returns show even larger long-horizon alphas, so the strategy's apparent outperformance is not an artifact of ignoring dividends."],"supporting_citations":[{"why":"Introduces the V/P strategy and the residual income valuation implementation that this paper builds on.","marker":"Frankel and Lee (1998)"},{"why":"Provides the linear information dynamics and residual income valuation framework used to estimate intrinsic value.","marker":"Ohlson (1995)"},{"why":"Supplies the empirical implementation of the residual income model and the treatment of 'other information' variables.","marker":"Dechow et al. (1999)"},{"why":"Basis for the simultaneous estimation of forecasting and valuation equations in a system, a key step in the paper's V/P construction.","marker":"Barth et al. (2005)"},{"why":"Defines the five-factor model whose inability to explain all V/P excess returns is the paper's central negative result.","marker":"Fama and French (2015)"},{"why":"Motivates the inclusion of profitability and investment factors in the benchmark asset pricing models.","marker":"Hou et al. (2015)"},{"why":"Frames the mispricing versus risk debate for the V/P anomaly that the paper extends to longer horizons.","marker":"Ali et al. (2003)"},{"why":"Earlier evidence that the Fama–French three-factor model cannot explain V/P returns, which the paper generalizes to longer holding periods and the five-factor model.","marker":"Hwang and Lee (2013)"},{"why":"Provides the GRS statistic used to test whether factor-model alphas are jointly zero across V/P portfolios.","marker":"Gibbons et al. (1989)"},{"why":"Shows the fundamental-to-market ratio subsumes book-to-market, providing context for why V/P is a distinct and powerful value signal.","marker":"Goncalves and Leonard (2023)"}],"fun_headline_variants":["High V/P stocks still beat risk models by year three","V/P ratio predicts returns for three years, US data","Value-to-price signal persists beyond year one","Mispricing lingers: V/P stocks earn alpha in years 2-3","V/P ratio edge lasts three years despite risk controls"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The significance of the long-horizon alphas relies on treating monthly returns from overlapping three-year holding periods as statistically independent, even though those returns are mechanically correlated because the same stock cohorts appear in many consecutive months.","fun_headline_variants_meta":{"raw":{"variants":["High V/P stocks still beat risk models by year three","V/P ratio predicts returns for three years, US data","Value-to-price signal persists beyond year one","Mispricing lingers: V/P stocks earn alpha in years 2-3","V/P ratio edge lasts three years despite risk controls"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000206,"raw_usage":{"total_tokens":1360,"prompt_tokens":871,"completion_tokens":489,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":487,"completion_tokens_details":{"reasoning_tokens":406}},"tokens_in":487,"tokens_out":489,"duration_ms":5974,"temperature":1.0,"reasoning_tokens":406,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T12:10:09.726706+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Re-estimate the five-factor alphas for the highest-V/P portfolio in years two and three using Newey–West or two-way clustered standard errors on the overlapping monthly return series; if the t-statistics fall below conventional thresholds, the paper's claim that factors cannot explain all variation loses its statistical support.","supporting_citations":[],"review_version":1}