{"id":"37c45e6f-7180-4700-bed8-697ce71e6676","arxiv_id":"2411.15228","paper_version":1,"verdict":"REJECT","confidence":"HIGH","novelty_score":3.0,"correctness_risk":"high","formal_verification":"none","parameter_count":6,"one_line_summary":"An auto-ARIMA analysis of Banda Aceh rice prices selects MA(5) models whose forecasts dip for five days and then hold constant, but that plateau is a mathematical property of MA(5), and the models are non-invertible.","lead":"This paper fits an ARIMA(0,0,5) model to six daily rice price series in Banda Aceh and forecasts a short decline followed by a flat price for the rest of 2023. The flat forecast is not an empirical finding; it is built into the MA(5) model, and the fitted coefficients are not valid.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The claim that ARIMA(0,0,5) is the best model for all six rice-price series is unsupported: the paper provides no stationarity test, the reported MA coefficients indicate non-invertible models, and the flat forecast is a mathematical property of an MA(5) with d=0.","rationale":"The reader's verdict is REJECT with high confidence, and I agree. The reader's weakest assumption -- that the series are stationary constant-mean processes suitable for a pure MA(5) -- is exactly the load-bearing point. My stress-test adds specificity: the reported equations (7)-(12) are numerically non-invertible, which is a red flag that the model is absorbing non-stationarity, and the paper provides no stationarity diagnostics. The forecast tables are also internally consistent with an MA(5) (the constant term after lag 5 equals the 'flat' forecast), so the headline pattern is purely structural. No ad hominem: the issue is methodological. A single ADF test on each series would settle whether the d=0 assumption holds; if the series are stationary, the concern is largely mitigated, but the paper provides no such evidence. Therefore the rejection stands unchanged.","tokens_in":12533,"tokens_out":4095,"duration_ms":40827,"concrete_test":"Run an Augmented Dickey-Fuller test (with trend) on each of the six imputed price series used in the paper, using the same daily data from PIHPS Nasional (or the values reproduced in Tables 3-4). If any series fails to reject the unit-root null at the 5% level, then d=0 is invalid, ARIMA(0,0,5) cannot be the best model, and the flat forecasts are an artifact of an incorrect differencing choice.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim is that each rice-price series is well described by a pure MA(5) in levels, so that d=0 and no trend or seasonal term is needed. This is the load-bearing premise: if any series has a unit root, a level, or a trend, then ARIMA(0,0,5) is misspecified and the forecasts in Tables 5-8 are not valid. The paper never reports an Augmented Dickey-Fuller, KPSS, or any other stationarity test. The data themselves argue against d=0: Table 3 shows BKB1 at 9850-10000 IDR in early 2020, while Table 4 shows 11600 IDR in August 2023, a clear long-run level shift. The reported equations (7)-(12) exacerbate the concern: all MA coefficients are positive and large (e.g., BKB1: 1+2.08B+2.74B^2+2.52B^3+1.56B^4+0.58B^5), with several coefficients exceeding 1 in absolute value, violating invertibility and suggesting that the model is trying to absorb non-stationarity through a non-invertible moving-average polynomial. Furthermore, any MA(5) forecast converges to the constant term after five steps, so the 'decline then constant' headline is an automatic consequence of the model choice, not a data-driven insight. The model-selection evidence (AIC/AICC/BIC comparisons, or auto.arima output) is absent, and the residual testing is only provided as unreadable figures with no Ljung-Box statistic or residual autocorrelation plot. In short, the central claim rests entirely on an unverified and implausible stationarity assumption.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper analyzes daily rice prices for six qualities in Banda Aceh from December 31, 2019 to August 31, 2023, applies last-observation-carried-forward imputation to missing values, and uses auto-ARIMA to select forecasting models. The authors claim that ARIMA(0,0,5) is the best model for all six series, present fitted equations with MA coefficients, and forecast a decline from September 1 to September 6, 2023 followed by constant prices through December 31, 2023. The manuscript also includes residual plots and forecast tables for the four-month horizon.","tokens_in":13084,"tokens_out":2749,"duration_ms":24530,"significance":"If fully supported, the paper would offer a concrete empirical forecasting result for a staple-food price series in a region where such studies are scarce. The authors' use of LOCF imputation and auto-ARIMA is methodologically conventional, and they do make an explicit, falsifiable forecast for a defined horizon. However, the central claims are not substantiated by the evidence presented: model selection criteria are never reported, the stationarity assumption is never tested, the fitted MA polynomials are implausible, and the headline 'decline then constant' pattern is a mathematical consequence of the chosen model rather than an empirical discovery. The paper therefore does not currently establish a reliable forecasting model or a substantive finding about rice prices.","major_comments":[{"comment":"The claim that ARIMA(0,0,5) is the 'best model' is unsupported because no AIC, AICC, or BIC values are reported anywhere in the manuscript. The abstract states that auto-ARIMA selects the best model based on one of these criteria, and Eqs. (4)-(6) define them, but no numerical values or comparisons among candidate models are given. Without this information, the reader cannot verify that ARIMA(0,0,5) outperforms other orders or that the choice is data-driven.","section":"Section III, below Eqs. (7)-(12)"},{"comment":"The fitted MA(5) models are not credible as stationary invertible representations of the data. For example, BKB1 in Eq. (7) has MA coefficients 2.08, 2.74, 2.52, 1.56, and 0.58, several of which exceed 1 in absolute value; similar patterns appear in all six equations. While individual coefficients greater than 1 do not automatically violate invertibility, the manuscript provides no check of the roots of the MA polynomial, and a standard auto-ARIMA routine would not normally return a non-invertible model. The reported coefficients suggest that the model is absorbing nonstationarity or level shifts through a non-invertible moving-average component, which undermines the interpretation of the equations as valid forecasting models.","section":"Section III, Eqs. (7)-(12)"},{"comment":"The load-bearing premise that each series is stationary in levels (d=0) is never tested. The paper reports no Augmented Dickey-Fuller test, KPSS test, or any other stationarity diagnostic. The data shown in Tables 1 and 3 themselves contradict this premise: BKB1 moves from 9850 IDR in early 2020 to 11600 IDR in August 2023, a clear long-run level shift, and similar shifts appear in the other series. If any of the six series contains a unit root or a trend, then ARIMA(0,0,5) in levels is misspecified and all forecasts in Tables 5-8 are invalid.","section":"Section III, pre-Eq. (7) and Tables 3-4"},{"comment":"The headline result that prices decline from September 1 to September 6 and then remain constant is a mathematical property of the fitted MA(5) model, not an empirical finding. For an MA(5) with a constant and no differencing, the h-step-ahead forecast converges to the constant after h=5, so the constant values repeated throughout Tables 6-8 are simply the estimated intercepts from Eqs. (7)-(12). The paper does not draw this connection and instead presents the plateau as a substantive forecast result, which is circular.","section":"Tables 5-8 and Figs. 7-12"},{"comment":"The residual diagnostics are inadequate to support the statement that 'the model is suitable for use in forecasting.' Figures 1-6 are plot images with no accompanying test statistics such as the Ljung-Box Q statistic, no residual ACF/PACF tables, and no numerical white-noise test results. Visual inspection of plots without objective criteria is not sufficient evidence of adequately modeled residuals, especially given the other specification concerns.","section":"Section III, Figs. 1-6"}],"minor_comments":[{"comment":"The notation in Eq. (1) is inconsistent: the text describes A_i-e as a recursive lag search, but the equation uses A_{i-a} without defining a; the manuscript should align notation and clarify the imputation rule.","section":"Section II, Eq. (1)"},{"comment":"The phrase 'c is konstanta' mixes Indonesian and English; it should read 'c is a constant.' Similar language issues appear elsewhere (e.g., 'Tabel' for 'Table' in Tables 6 and 8).","section":"Section II, text after Eq. (3)"},{"comment":"Figure 6 is captioned 'Residuals testing from BKS1 variable,' but Eq. (12) and the surrounding text indicate it should be for BKS2; the label should be corrected.","section":"Section III, Eqs. (7)-(12) and Fig. 6"},{"comment":"Several numerical inconsistencies appear between the text and tables: the text reports BKS1 September 4 as 12358 IDR and September 5 as 12419 IDR while Table 5 shows 12359 and 12420; for BKS2 the text lists an extra '11704 IDR' and gives a seven-value sequence for a six-day period. These discrepancies should be reconciled.","section":"Section III, text around Tables 5 and 6"},{"comment":"The conclusion states that the super quality rice II forecast range is 'between 11388 IDR and 12761 IDR' but refers to it as 'super quality rice I,' repeating the earlier mislabeling; this should be corrected.","section":"Section IV, Conclusions"}],"recommendation":"reject","confidential_remarks":"The manuscript appears to be a lightly edited version of a paper published in a non-archival journal (IJCER), and its presentation falls well below the standard of a serious statistics or econometrics journal. The central claims are unsupported by the reported evidence: model selection values are absent, stationarity is untested, the fitted coefficients are implausible, and the main forecast pattern is an artifact of the model structure. These issues are load-bearing rather than cosmetic, so rejection is appropriate. If the authors redo the analysis with proper diagnostics, model comparison, and out-of-sample validation, a substantially revised version might be considered, but the current manuscript does not provide a sound basis for publication."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this is a routine auto-ARIMA exercise on daily rice prices for six qualities in Banda Aceh, and the central claim—that ARIMA(0,0,5) is best and that prices fall then flatten—does not survive contact with the reported equations. The flat forecast is a built-in property of any MA(5) model, not a discovery about rice prices.\n\nWhat the paper does honestly: it picks a concrete local problem, uses a real data source (PIHPS Nasional), documents missing values, and shows its full forecast tables. That transparency is good, and the local context is genuinely under-studied; I can believe no one has fit these six series before. The LOCF imputation is described exactly, and the authors acknowledge that differencing should be tried later.\n\nThe soft spots are load-bearing, though. For every series, the reported MA coefficients are larger than 1 in absolute value (BKB1's first coefficient is 2.08), so the fitted model is non-invertible. That alone indicates misspecification, and the most plausible cause is non-stationarity: the data start around 9850 IDR and end near 11600 IDR, but the model uses d=0 with no trend or stationarity test. No AIC, AICC, or BIC values are reported, despite the abstract claiming selection by those criteria. The residual diagnostics are figures without test statistics; no Ljung-Box or normality test is given. And the forecast tables are just the MA(5) mean reversion: after five steps the predictions equal the constant, so Tables 5–8 are restatements of the estimated constants. The \"decline then constant\" headline is therefore an artifact of the model class.\n\nThat said, the paper is not a mess in every dimension. It is readable, the method section is standard, and the authors are open about their limitations. The problem is that the conclusion rests on unverified and implausible assumptions.\n\nWho could use this? Only as a cautionary example for a forecasting course, or as a baseline for a more careful reanalysis. I would not cite it. A serious editor should desk-reject it in its current form; the statistical errors are fundamental and would require redoing the entire analysis, not just patching a section. If the authors redo the study with proper stationarity testing and model selection, there is a modest applied paper in there.","headline":"A routine ARIMA application undercut by non-invertible models and missing model-selection evidence; the flat forecast is a mathematical artifact, not a finding.","tokens_in":13539,"tokens_out":2954,"would_cite":false,"duration_ms":26054,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":false},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["91B84","62-04"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper claims that daily rice prices in Banda Aceh after Covid-19 follow an ARIMA(0,0,5) model and forecasts a five-day drop followed by flat prices through December 2023.","keywords":["Missing value","imputation","LOCF","auto-ARIMA","rice quality","Banda Aceh","price forecasting","ARIMA"],"falsifier":"Compare the model's flat forecasts with actual daily rice prices for Banda Aceh from September 7 through December 31, 2023; any sustained upward or downward movement would contradict the claim. Running an Augmented Dickey-Fuller test on the imputed series would also check whether the d=0 (no differencing) choice is justified.","tokens_in":12374,"feed_emoji":"🍚","tokens_out":9553,"duration_ms":79748,"temperature":0.7,"pith_summary":"This paper tries to establish that for six qualities of rice sold in Banda Aceh's traditional market, a single model order—ARIMA(0,0,5)—best describes the daily price series after Covid-19, and that the fitted models forecast a short decline in early September 2023 and then constant prices until the end of the year. The claim matters because stable rice prices are a food-security concern in Indonesia, and a simple model that can be refitted as new data arrive would give market monitors a cheap forecasting tool. The paper arrives at this by filling missing daily prices with the last observed value (LOCF) and then letting auto-ARIMA choose the order by AIC/AICC/BIC. If true, the six reported equations would be the working forecasting equations for these qualities.","feed_headline":"ARIMA(0,0,5) selected for all six rice price forecasts","feed_subtitle":"After Covid-19, daily rice prices in Banda Aceh are forecast to dip for five days, then hold constant through 2023.","key_machinery":"The machinery is the ARIMA(0,0,5) model—a moving-average model of order five, written as $X_i = c + \\varepsilon_i + \\theta_1\\varepsilon_{i-1} + \\cdots + \\theta_5\\varepsilon_{i-5}$, where $X_i$ is the daily price, $c$ the constant, and $\\varepsilon$ the error. Because $d=0$, the model assumes the price series is already stationary and mean-reverting around a constant, with no trend or seasonality. The auto-ARIMA routine selects this order for all six qualities based on the AIC/AICC/BIC criteria, and the equations carry the forecast: the moving-average terms die out after five days, leaving the constant as the flat long-run prediction.","core_discovery":"The central discovery is that all six rice price series—lower quality I and II, medium quality I and II, super quality I and II—are best modeled as a pure moving-average process of order five, ARIMA(0,0,5), with no autoregressive terms and no differencing. Fitted to daily prices from the Banda Aceh traditional market between December 31, 2019 and August 31, 2023, the models produce forecasts that decline for the first six days of September 2023 and then become constant from September 7 through December 31, 2023. The paper reports the six fitted equations (7)–(12), one per quality, and uses the constants of those equations as the long-run flat forecast levels.","pith_inferences":["The d=0 assumption is never tested with stationarity diagnostics; if the series are actually trending or seasonal, a differenced or seasonal ARIMA would likely give different, non-flat forecasts.","The perfectly flat forecast after five days is a structural feature of the pure moving-average model, not an empirical prediction that prices will literally stop moving; real market data typically fluctuate.","Applying the same auto-ARIMA pipeline to realized prices from September–December 2023 would provide a direct out-of-sample check on whether the constant forecasts held.","The same five-day memory could be tested on other staples in Banda Aceh (e.g., cooking oil, sugar) to see whether the short-memory pattern is specific to rice or a general market feature."],"forward_implications":["If the ARIMA(0,0,5) model is right, price shocks in Banda Aceh's rice market fade within five days, after which the series returns to a constant level.","The forecast flat levels (e.g., 9995 IDR for lower quality I, 10423 IDR for lower quality II) are the model's predicted steady-state prices from September 7 to December 31, 2023.","The brief September 1–6 decline implies a one-time correction from August price levels, not a continuing trend.","Because the same model order fits all six qualities, the market appears to move together, so forecasts for one quality can cross-check the others.","The LOCF imputation creates a complete daily panel, which is what makes daily forecasts with no gaps possible."],"supporting_citations":[{"why":"Supplies the direct antecedent: ARIMA forecasting of medium-quality rice prices, which this paper extends to six qualities in Banda Aceh.","marker":"[10]"},{"why":"Provides the ARIMA modeling context and hybrid ARIMA approach that frame this paper's forecasting method.","marker":"[8]"},{"why":"Demonstrates ARIMA model selection via AIC, the same criterion the auto-ARIMA routine uses to pick ARIMA(0,0,5).","marker":"[12]"}],"fun_headline_variants":["ARIMA(0,0,5) foresees 5-day rice dip, then steady","Banda Aceh rice: brief drop in Sept, then flat","Post-COVID rice forecast: short dip, long flat","Auto-ARIMA picks MA(5) for all rice qualities","Rice prices to dip five days, then hold till year-end"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The argument assumes that rice prices in Banda Aceh are stationary around a fixed average, with no trend or seasonal pattern, so a pure moving-average model with zero differencing captures the data, and this assumption is asserted rather than demonstrated with stationarity tests.","fun_headline_variants_meta":{"raw":{"variants":["ARIMA(0,0,5) foresees 5-day rice dip, then steady","Banda Aceh rice: brief drop in Sept, then flat","Post-COVID rice forecast: short dip, long flat","Auto-ARIMA picks MA(5) for all rice qualities","Rice prices to dip five days, then hold till year-end"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001116,"raw_usage":{"total_tokens":4632,"prompt_tokens":913,"completion_tokens":3719,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":529,"completion_tokens_details":{"reasoning_tokens":3625}},"tokens_in":529,"tokens_out":3719,"duration_ms":26110,"temperature":1.0,"reasoning_tokens":3625,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T15:30:53.093642+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compare the model's flat forecasts with actual daily rice prices for Banda Aceh from September 7 through December 31, 2023; any sustained upward or downward movement would contradict the claim. Running an Augmented Dickey-Fuller test on the imputed series would also check whether the d=0 (no differencing) choice is justified.","supporting_citations":[{"cited_title":"Arima Model for Forecasting the Price of Medium Quality Rice to Anticipate Price Fluctuations,","cited_arxiv_id":null,"evidence_quote":"Supplies the direct antecedent: ARIMA forecasting of medium-quality rice prices, which this paper extends to six qualities in Banda Aceh."},{"cited_title":"ARIMA -AdaBoost hybrid approach for product quality prediction in adva nced transformer manufacturing,","cited_arxiv_id":null,"evidence_quote":"Provides the ARIMA modeling context and hybrid ARIMA approach that frame this paper's forecasting method."}],"review_version":1}