{"id":"4dc68652-19a8-4002-aaf9-92c0ff716557","arxiv_id":"2607.29162","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"In Japan's Fundinno equity crowdfunding, same-prefecture users are 6.7 percentage points more likely to invest after viewing a non-Tokyo campaign page, versus a 10% baseline and a 0.9 pp adjacent-prefecture premium.","lead":"Using detailed user-campaign logs from Japan's largest equity-crowdfunding platform, the paper finds that investors outside the Tokyo area are substantially more likely to fund campaigns from their own prefecture even after they have opened the campaign page. The result separates geography-driven attention from what happens after access, showing that online platforms can broaden reach without making investment demand geographically neutral.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Same-prefecture view-to-investment premium may be inflated by issuer outreach or prior ties that also drive page views; timing of first view (preview vs. investment phase) can adjudicate.","rationale":"The reader's weakest assumption is that the view-to-investment margin measures post-access demand rather than unobserved same-prefecture-specific outreach or prior ties. I agree that this is the most load-bearing concern: the paper's novelty rests on separating discovery from post-access behavior, and the connected-investor screen is demonstrably incomplete. My proposed test—using the preview phase and view timing—directly targets this assumption and would either confirm the margin's interpretability or expose a severe confound. I did not find a more fundamental technical flaw: the two-way fixed-effects design is standard, the support tables are reassuring, and the paper is appropriately hedged about motives. The lack of a replication package is a concern, but it does not change the verdict from conditional, since the empirical pattern is carefully presented. Thus the verdict remains CONDITIONAL, with the concrete timing test as the natural path to strengthening or rejecting the interpretation.","tokens_in":16790,"tokens_out":9406,"duration_ms":93101,"concrete_test":"Re-estimate the non-TMA view-to-investment specification (Equation 2) splitting the sample by the timing of the first detail-page view. Fundinno campaigns have a preview phase during which the page can be read but no investment is possible (Section 2.3). Classify each view-to-investment observation by whether the first view occurred during the preview phase (before the investment window opened) or during the investment phase; among the latter, flag same-day first-view-and-investment pairs. If the 0.067 same-prefecture coefficient is concentrated in same-day view-invest sequences and absent among preview-phase viewers, the outreach/prior-ties channel is supported. If the coefficient is stable (or even larger) for preview-phase viewers—who had time to read the page before deciding—then the post-access interpretation is strengthened. A secondary check: use the number of detail-page views (e","verdict_should_be":"UNCHANGED","load_bearing_attack":"The core estimate (Table 6: non-TMA θ_I,same = 0.067, SE 0.007) is conditional on a recorded detail-page view. The paper interprets this as evidence that the local premium 'remains after campaign-page access' and is 'difficult to explain by discovery alone.' This interpretation requires that the view event marks a meaningful stage after which further local advantage is not merely a reflection of pre-existing connections. The connected-investor screen (Section 5.4) is narrow: it excludes only users who registered within 21 days, invested in one campaign, and viewed fewer than five pages. It cannot remove more common prior ties or direct issuer outreach to acquaintances, which are likely correlated with same-prefecture status. If such outreach raises both the probability of viewing and the probability of investing, conditioning on view selects on unobserved pair-specific 'activation.' The 0.067 coefficient could then overstate a post-access local premium and the same-versus-adjacent gap (0.067 vs 0.009) could partly reflect that adjacent-prefecture users are far less likely to be personally contacted. There is also a collider-bias concern: local affects viewing (e.g., non-TMA listing-to-view same-prefecture coefficient 0.027, and 0.120 with display), so conditioning on view can create selection bias in the investment margin even if local has no causal effect after access. The paper is careful in places, but the central claim that the premium is not purely a discovery effect relies on the view-to-investment comparison being informative about post-access behavior. Without additional timing or intensity information, the 6.7-point premium could be an artifact of who views, not of what happens after viewing.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper uses proprietary two-stage (listing exposure, detail-page view, investment) user-campaign records from Fundinno, Japan's dominant equity-based crowdfunding platform, to decompose the geography of online investment into a discovery margin (listing-to-view) and a post-access margin (view-to-investment). With user and campaign fixed effects on 527 campaigns, it finds that for campaigns outside the Tokyo metropolitan area (non-TMA), same-prefecture viewers are 6.7 percentage points more likely to make a positive net investment after viewing the campaign detail page, relative to other-prefecture viewers whose rate is about 10 percent. The adjacent-prefecture difference is only 0.9 percentage points, and the same-prefecture premium for TMA campaigns is 0.6 percentage points. Logit average marginal effects are reported as nonlinear checks; for the non-TMA investment margin the logit AME is 9.2 percentage points. A robustness exercise excluding narrowly-defined likely connected investors lowers the non-TMA same-prefecture estimate to 5.1 percentage points. The paper interprets the pattern as evidence that the local premium is not purely a discovery effect and is not a smooth proximity gradient, while explicitly acknowledging that the data do not identify the underlying motive.","tokens_in":17099,"tokens_out":6475,"duration_ms":67214,"significance":"If the central result holds, the paper makes a meaningful contribution to the literature on geography in online financial markets. Its strengths are the unusually rich platform data linking exposure, viewing, investment, and geography for nearly the entire Japanese ECF market; the clean separation of two conditional margins; the use of two-way fixed effects with two-way clustered standard errors; the inclusion of logit checks; and the transparent support tables in the appendix. The paper is appropriately cautious in labeling its estimates as descriptive and in stating that prior ties, issuer outreach, and home-prefecture support motives remain possible channels. The main limitation is that the post-access interpretation relies on the view-to-investment contrast being informative about investment demand after access, rather than about selection into viewing; the paper does not exploit the preview/investment phase timing to address this directly.","major_comments":[{"comment":"The view-to-investment coefficient theta^I,same_non-TMA = 0.067 is conditional on a recorded detail-page view. Because same-prefecture status strongly affects the earlier margin (Table 6, listing-to-view: 0.027, with an additional 0.093 when geographic information is displayed), the investment-margin contrast is identified on a selected sample. The paper acknowledges this in §4.3, but the preview/investment phase described in §2.3 provides a feasible test: report specifications that split by whether the user's first detail-page view occurred during the preview phase (when investment is impossible) or during the investment phase. If the 6.7pp premium is present among preview-phase first viewers, the interpretation that the premium appears 'after campaign-page access' is strengthened; if the premium is absent there, the result may largely reflect same-prefecture users viewing once they are","section":"§5.1, Table 6; §2.3"},{"comment":"The connected-investor exclusion removes only users who registered within 21 days, invested in one campaign, and viewed fewer than five pages. This screen does not remove more common prior ties or direct issuer outreach, which are plausibly correlated with same-prefecture status. The estimate falls from 0.067 to 0.051, but the residual could still be driven by such ties. I recommend either stronger screens (e.g., excluding users with any history of same-prefecture page views or investments, or using registration timing relative to campaign start) or a formal sensitivity/bounds analysis, and a more cautious statement that issuer outreach/prior ties are not the primary driver.","section":"§5.4, Table 8"}],"minor_comments":[{"comment":"The logit AME for non-TMA same-prefecture view-to-investment is 0.092 versus the LPM 0.067. This is a non-trivial gap; please discuss whether it reflects attrition in the fixed-effect logit sample, nonlinearity, or differences in the averaging method, and present the main magnitude as a range if appropriate.","section":"Table 6"},{"comment":"Please define 'qualifying listing-page exposure' and 'first observed detail-page view' precisely, including how ties or orderings between exposure and view are handled when constructing the samples.","section":"§3.4"},{"comment":"The statement that LPM coefficients are 'directly interpretable as percentage-point differences' is fine, but the computation of the logit average marginal effects (the baseline over which averaging is performed) should be documented in the table notes or text.","section":"§4.5"},{"comment":"The two Japan Securities Dealers Association (2026) entries are both cited as 2026; please distinguish them by adding months or report numbers, as in the text.","section":"References"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is empirically careful and the data are unusually good. However, the central interpretation depends on the view-to-investment contrast being interpreted as a post-access margin, and the paper does not use the preview/investment timing that it itself describes in §2.3. If the authors provide that split and it supports the premium, I would move to accept; as it stands, this is a major revision."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The paper is worth a serious look. What is actually new is that it uses Fundinno's click-level records to separate listing exposure, campaign-page views, and investment, and then shows that for non-TMA campaigns, same-prefecture users are 6.7 percentage points more likely to invest after viewing a page. That is a sharper fact than the usual 'realized investments are local' result found in prior work. The adjacent-prefecture contrast is only 0.9 points, and the TMA/non-TMA split is sensible given how prefectural boundaries work in Japan. The fixed-effects specifications, logit checks, support tables, and the connected-investor screen are all done competently, and the main text is appropriately hedged: the authors repeatedly say they cannot identify the underlying motive.\n\nSoft spots: the data are proprietary and there is no replication archive, so independent checks will be hard. The connected-investor screen is narrow and does not remove more common prior ties or issuer outreach. The more serious issue is the stress-test's collider concern: since local status also affects the probability of viewing (0.027 for non-TMA), conditioning on view can select on unobserved pair-specific activation, potentially biasing the view-to-investment estimate even if local has no causal effect after access. The paper acknowledges this in Section 4.3, but the abstract's 'difficult to explain by discovery alone' pushes a bit past what the design identifies. The view event is also coarse—no timing or page-intensity information. A clean next step would be to compare investors whose first view occurred in the preview phase versus the investment phase, or to use listing-stage display as an instrument for access.\n\nNone of this kills the paper. The 5.1-point result after excluding likely connected investors is still meaningful, and the adjacent-prefecture gap is harder to explain by pure selection. But the headline should be read as descriptive: same-prefecture viewers are more likely to invest post-access in non-TMA campaigns, and the mechanism remains open. The paper is a solid empirical contribution to the geography-of-finance literature, and the main caveats are addressable. I would send it to a serious referee, and I would bring it to a reading group if you want a good discussion of conditioning-on-intermediate-variable bias.","headline":"A careful stage-level decomposition of local bias in Japanese equity crowdfunding, with a credible same-prefecture post-view premium that survives a connected-investor screen, though the causal interpretation is constrained by selection into page views.","tokens_in":17606,"tokens_out":2237,"would_cite":true,"duration_ms":24792,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Same-prefecture investors 6.7 points likelier to fund regional deals","keywords":["geography","equity crowdfunding","local bias","home bias","Japan","investor behavior","fixed effects","Fundinno"],"falsifier":"Compute the view-to-investment premium for non-TMA campaigns separately for users who reached the campaign page directly from an external link (bypassing the listing page). If the 6.7-point same-prefecture premium collapsed to zero in that direct-access subsample, the claim that the premium represents post-access demand rather than residual discovery or attention would be falsified.","tokens_in":16661,"feed_emoji":"🏘️","tokens_out":6802,"duration_ms":55881,"temperature":0.7,"pith_summary":"This paper asks whether online access makes investment geography a thing of the past. Using user-campaign records from Japan's dominant equity-crowdfunding platform, Fundinno, it traces each user from listing exposure to campaign-page view to investment, and compares same-prefecture, adjacent-prefecture, and other-prefecture users. The central finding: for campaigns outside the Tokyo metropolitan area, viewers from the issuer's own prefecture are 6.7 percentage points more likely to invest after viewing the campaign page, relative to the roughly 10 percent investment rate among other-prefecture viewers. The adjacent-prefecture premium is only 0.9 points, and the same-prefecture premium is small for Tokyo-area campaigns. This shows that the local premium is not only a discovery effect: it persists after users have opened the campaign's detailed information page.","feed_headline":"Same-prefecture investors 6.7 points likelier to fund regional deals","feed_subtitle":"For non-Tokyo campaigns the home-field edge persists after viewing the pitch page; proximity alone can't explain it.","key_machinery":"The identifying device is the two-stage user-campaign funnel recorded by the platform: listing exposure -> campaign detail-page view -> investment. Using two-way fixed effects (user and campaign), the paper examines two conditional margins — listing-to-view (likely discovery) and view-to-investment (post-access demand). The key comparison is the same-prefecture coefficient in the view-to-investment regression for non-TMA campaigns, benchmarked against adjacent-prefecture and TMA coefficients, so the same-prefecture locality claim rests on the contrast across those groups.","core_discovery":"The paper establishes that a same-prefecture premium in equity-crowdfunding investment survives conditioning on campaign-page access. For non-TMA campaigns, same-prefecture viewers have a 6.7 percentage-point higher probability of positive net investment relative to other-prefecture viewers, against a base rate near 10 percent. The adjacent-prefecture difference is only 0.9 points and the same-prefecture difference for TMA campaigns is 0.6 points, so the premium is concentrated in same-prefecture pairs outside the Tokyo metropolitan area. Because the premium appears after users view the detailed campaign page, it cannot be explained by discovery alone; the small adjacent-prefecture gap also","pith_inferences":["A design experiment that randomizes the display of the issuer's prefecture on the listing page could isolate whether the viewing-margin premium is a cue effect; the paper's display evidence is descriptive because display is not assigned.","Applying the same funnel decomposition to other online finance platforms (green bonds, real estate crowdfunding) would test whether the post-access home premium is specific to equity or general to digital capital allocation.","The authors remain agnostic about motive; if the premium is nonpecuniary home support, then even perfect information and friction-free platforms would not flatten the geographic pattern, whereas an information channel would respond to enhanced disclosure.","For policymakers, the results imply that promoting online equity finance in rural areas may not by itself reduce reliance on local capital; supplementary policies might target outside investors or inter-regional networks."],"forward_implications":["If the premium truly operates after campaign-page access, online platforms can widen reach without making capital allocation geographically neutral.","Regional (non-TMA) issuers may continue to rely on same-prefecture households even when listed on a national platform.","The modest adjacent-prefecture gap implies frictions or preferences are bound to prefectural boundaries, not to continuous distance, at least at this level of aggregation.","The small TMA same-prefecture premium suggests that 'same prefecture' captures distinct local context outside the metropolitan core, not simply the administrative boundary.","If the underlying motive were identified, the result could speak to whether the platform's local information environment or purely nonpecuniary home-support drives the effect."],"fun_headline_variants":["Home-prefecture edge persists after crowdfunding pitch views","Same-prefecture investors 6.7 pts likelier after viewing pitch","Local funding gap survives page views in Japanese crowdfunding","Equity crowdfunding: local bias remains after campaign access","Regional crowdfunding still favors same-prefecture investors"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The interpretation hinges on the claim that the view-to-investment margin isolates demand generated after the page view, and not pre-existing same-prefecture connections (issuer outreach, prior ties) that survive the connected-investor screen.","fun_headline_variants_meta":{"raw":{"variants":["Home-prefecture edge persists after crowdfunding pitch views","Same-prefecture investors 6.7 pts likelier after viewing pitch","Local funding gap survives page views in Japanese crowdfunding","Equity crowdfunding: local bias remains after campaign access","Regional crowdfunding still favors same-prefecture investors"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000173,"raw_usage":{"total_tokens":1117,"prompt_tokens":749,"completion_tokens":368,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":493,"completion_tokens_details":{"reasoning_tokens":288}},"tokens_in":493,"tokens_out":368,"duration_ms":3842,"temperature":1.0,"reasoning_tokens":288,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-03T12:26:26.032130+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute the view-to-investment premium for non-TMA campaigns separately for users who reached the campaign page directly from an external link (bypassing the listing page). If the 6.7-point same-prefecture premium collapsed to zero in that direct-access subsample, the claim that the premium represents post-access demand rather than residual discovery or attention would be falsified.","supporting_citations":[],"review_version":1}