Pith. sign in

REVIEW 2 major objections 4 minor 65 references

AI Financial Advice: Supply, Demand, and Life Cycle Implications

T0 review · 2 major / 4 minor · reviewed 2026-08-05 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read Claim: following a chatbot's financial advice would move most households toward textbook life cycle behavior — broader equity, bigger buffers — and a third of the gender gap in advice comes from the model reading a gender label.

desk verdict A serious paper whose qualitative findings are likely right; the one-third supply-side gender estimate is the one number I would not bet on until the translation step is validated on the label experiment itself. read the letter →

arxiv 2608.01607 v1 pith:VTOCQLWT submitted 2026-08-03 econ.GN q-fin.ECq-fin.GNq-fin.PM

Taha Choukhmane , Tim de Silva , Weidong Lin , Matthew Akuzawa This is my paper
classification econ.GNq-fin.ECq-fin.GNq-fin.PM MSC 91B4291G10
keywords LLMfinancialadvicelifecycleportfoliochoicehouseholdfinancepromptheterogeneitygendergapindemandversussupplyconsumptionsmoothingdrift
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper tries to establish that the saving, spending, and investing advice of a leading chatbot, followed literally each year in a simulated adult life, would move most U.S. households closer to what standard life cycle theory prescribes: near-universal ownership of diversified equity funds, equity shares that decline after age 45, and savings buffers above $10,000 by age 30. It reaches this by a new method — eliciting realistic prompts from a representative survey sample, feeding them to the model with each simulated individual's dollar amounts swapped in, and using a second model to translate the textual advice into quantitative choices within a calibrated life cycle model. The paper also argues the advice departs from the theory's finer points: it implies an annual discount factor above one, relies on round-number and 4%-withdrawal heuristics, cuts consumption sharply after job loss, and lets portfolios drift passively with returns. Its most distinctive claim is the gender decomposition: about two-thirds of the gender gap in recommended equity shares comes from men and women writing different prompts, and one-third from the model giving different advice to identical prompts labeled 'I am a man' versus 'I am a woman.' A sympathetic reader cares because, if right, cheap AI advice becomes a countervailing force to documented household finance frictions — while carrying a small but measurable supply-side gender bias that regulators and model builders can act on.

What carries the argument

The engine is a three-stage pipeline: a survey collects three free-text prompts per respondent (finance description, spending question, investing question) plus demographics, literacy, and AI experience; a calibrated life cycle model supplies stochastic income, unemployment, mortality, taxes, and four asset classes; and each simulated year a prompt from a similar-age/income/employment respondent is drawn, dollar amounts are replaced with the simulated agent's states, the advice model (GPT-5.2) answers in text, and a second model (GPT-5 Mini) deterministically converts text into dollar consumption and asset contributions. The decomposition's engine is the randomized-label experiment: gender-n

What would settle it

Replace the text-to-numbers translation step with direct structured output from the advice model for every demographic subgroup and the full life cycle (the paper only does this for GPT-5.2 on aggregate profiles), or have human coders re-encode a sample of the same textual advice into the same dollar allocations. If the direct-JSON variant changes the two-thirds/one-third gender split, the post-45 equity decline, or the 4–6% wealth gaps, then the translation step — not the advice model — is producing the paper's economics.

Watch

Extended reading notes

Core claim

The central claim: taken literally, LLM saving-and-investing advice would move most people toward standard life cycle theory — near-universal diversified equity ownership, equity shares declining after 45, savings buffers above $10,000 by age 30 — while departing on finer points (implied discount factor above one, the 4% withdrawal rule, sharp consumption drops after job loss, passive portfolio drift). Advice also differs by who asks: women's, low-literacy, and non-AI-user prompts yield 4–6% lower simulated wealth at age 60. Randomized gender labels split the equity-advice gap into two-thirds demand (different prompt content) and one-third supply (different advice to identical prompts labele

Load-bearing premise

All the quantitative results — the simulated life cycles, the 4–6% wealth gaps between groups, and the two-thirds/one-third gender split — rest on a second AI model reliably converting free-text advice into dollar amounts for spending, saving, and each asset class, with any conversion errors unrelated to age, gender, or prompt content.

Editorial extensions

If this is right

  • If followed, the advice would counter documented frictions: simulated households reach near-universal stock participation, equity shares that fall after 45, and buffers above $10,000 by age 30, unlike their self-reported status quo.
  • The advice premium is partly in users' hands: the same model buys a 1.50 percentage-point lower recommended diversified equity share for a woman-written, woman-labeled prompt, and a structured researcher-designed prompt fixes most heuristics and consumption smoothing but not portfolio inertia.
  • Group differences compound over working life: simulated wealth at age 60 is roughly 4–6% higher under prompts from men, high-literacy respondents, and prior AI users, so AI advice could widen existing wealth gaps even while improving average outcomes.
  • The one-third supply-side gender effect is a concrete audit target: identical prompt text consistently receives lower equity recommendations when prefixed 'I am a woman,' a pattern a model developer could measure and mitigate directly.
  • The results characterize the effect of following advice, not of receiving it; actual behavior change depends on adherence, which the paper does not measure.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The same label-randomization design could quantify supply-side effects of other personal markers the model might use as risk-tolerance proxies — age, occupation, parenthood, or dialect — and the one-third gender share suggests such proxies are active in current models.
  • Since real households rarely follow advice precisely, the simulated 4–6% wealth gaps are best read as an upper bound on AI advice's redistributive impact, not a point prediction.
  • The stochasticity measurement (median 6.5 percentage-point equity-share variation across repeated queries, with translation contributing only 1.6 points) identifies advice generation, not extraction, as the noise source a monitoring regime should track.
  • The pipeline is re-runnable: as new models ship, repeating the survey-and-simulation pass yields a time series against life cycle diagnostics — effectively a public benchmark for financial advice quality that does not exist today.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

2 major / 4 minor

Summary. The paper develops a three-step method to quantify LLM financial advice: (i) a Prolific survey elicits three free-text prompts (financial situation, spending, investment) from 952 U.S. adults; (ii) a life cycle model calibrated to U.S. data provides the economic environment and a normative benchmark; (iii) simulated individuals are matched to prompts in age/income/employment buckets, their state variables are inserted into the prompt text, GPT-5.2 generates advice, and GPT-5 Mini translates the advice into dollar consumption/saving/asset-allocation choices. The method is applied to GPT-5.2, with robustness to Gemini 3 Flash and GPT-5.6 Terra. Main findings: following LLM advice would move respondents toward broader equity participation, age-declining equity shares, and larger savings buffers; advice departs from the model in high patience, heuristics, imperfect consumption smoothing, and passive drift; and advice varies by gender, literacy, and AI experience. A randomized-label experiment attributes roughly two-thirds of the gender gap in diversified equity recommendations to demand (prompt content) and one-third to supply (gender label).

Significance. The paper has substantial strengths: a rich survey eliciting genuine user prompts, a serious life cycle model, standard errors throughout, repeated-query quantification of LLM stochasticity, robustness across three models, and a direct-JSON robustness pipeline. The randomized-label design is a genuine contribution to separating demand and supply in AI advice. If the quantitative translation step is valid, the paper would be an important benchmark for AI household-finance advice. The central caveat is that all quantitative outcomes, including the headline gender decomposition, depend on an unvalidated second-LLM translation step; the only direct validation covers aggregate profiles, not the label experiment. This is the main load-bearing weakness.

major comments (2)
  1. [1.3, Step 4; A.5.2; Table 3; E17; E12] Every quantitative outcome in the paper, including the gender decomposition, is produced by a second LLM (GPT-5 Mini) translating free-text advice using hand-written deterministic rules. The paper validates this translation step only through the direct-JSON robustness check (Section 3.4, Figure E12), which replicates aggregate consumption and equity-share profiles for GPT-5.2; it does not re-estimate the coefficients in Table 3. The label coefficient βS = -0.54pp is the key supply-side estimate, and the median translation-only stochasticity for equity shares is 1.6pp (Figure E17), roughly three times that coefficient. A systematic extraction error—for example, the STOCK-SPLIT DEFAULT or the D+I mapping of 'stocks' interacting with wording that differs by gender label—could create or mask the label effect. Please provide a direct-JSON replication of the Table 3 label experiment, or valida
  2. [1.1; Table E1] The abstract and Section 1.1 describe the sample as 'representative' or 'demographically balanced,' but Table E1 shows the Prolific sample over-represents the unemployed (15% vs 3% in CPS) and under-represents those 70-79 (8% vs 11%) and 80+ (0% vs 5%), with no reweighting or sensitivity analysis. Because the prompt pool is the input to all life cycle simulations and to the demand-side comparisons, the level and external validity of the simulated profiles may depend on this composition. At minimum, add a weighted or reweighted robustness exercise, and qualify the 'representative' claim in the abstract.
minor comments (4)
  1. [4.2; Table 3] The two-thirds/one-third decomposition is estimated on the 83% of prompts that do not explicitly mention gender. State this caveat in the abstract or conclusion, and provide a delta-method confidence interval for the ratio βD/(βD+βS), which is currently reported without uncertainty.
  2. [Appendix D; Table 2] The SMM estimates of β and γ are reported as points on a grid without standard errors or a confidence set. Since the claim of 'unusually high patience' relies on β>1, please report the grid sensitivity or a bootstrap/confidence region.
  3. [Figure E17] The 'Translation only' exercise repeats the translation of fixed advice five times. Clarify in the note that the 1.6pp median is across repetitions and does not capture systematic bias; otherwise readers may infer that stochasticity is the only concern.
  4. [1.2; 2.2] The term 'equity share' is defined in Section 1.2 but used in figures and tables before being reintroduced. Consider defining it at first use in Section 2.2, especially since Figure 5's middle panel reports conditional-on-participation averages.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: LLM advice is measured externally; SMM parameters are summaries, not inputs.

full rationale

The paper's quantitative backbone is an external measurement pipeline: survey respondents write prompts, GPT-5.2 produces textual advice, GPT-5 Mini translates that advice into choices using deterministic rules, and the life cycle model then accumulates the consequences. The life cycle model is calibrated with external data (SIPP, CRSP, SSA) and builds on the authors' prior Choukhmane and de Silva (2026) model, but that benchmark is not an input that generates the LLM advice; it is a comparison object. The claims about moving toward life cycle theory are comparisons between measured LLM recommendations and observed behavior or model benchmarks, not derivations from the benchmarks. The SMM estimates of beta and gamma are explicitly fitted to the LLM-generated moments and are used only as a summary of how the advice departs from the model; they are not then used to produce the advice or the headline facts. The gender supply/demand decomposition is identified by randomized gender labels in regression (1), so the label coefficient is an experimental contrast, not a definitional identity. The translation step's lack of validation for the gender-label experiment is a measurement-validity concern, not circularity: nothing in the paper equates the extracted choices to the extraction rules by construction, and the translation rules do not embed the gender coefficients. Self-citations appear in the calibration of the life cycle model, but the cited parameters come from external datasets and the central LLM-advice results do not reduce to that citation. Overall, the derivation chain is self-contained with respect to its central claims.

Assumptions & free parameters 3 free parameters · 6 assumptions · 0 invented entities

The central claims rest on a calibrated simulation model and a two-LLM pipeline. None of the paper's assumptions are formalized or machine-checked, and several ad hoc choices (bucket cutoffs, household scaling, translation rules) are hand-selected by the authors. The most important parameters are estimated externally (SIPP, CRSP, SSA), which is appropriate, but the translation heuristics are internal and unvalidated. No new physical entities are introduced.

free parameters (3)
  • Prompt bucket cutoffs (age and income terciles, unemployed halves, single retired bucket) = heuristic thresholds from the survey sample (Appendix A.3)
    Determines which respondent's prompt is assigned to each simulated individual; coarse bins mean a simulated 60-year-old may draw a prompt authored by a 45-year-old, affecting the age profile and heterogeneity estimates.
  • Household scale-up factor for inserted income/wealth = double unless more than two working adults are mentioned (Appendix A.1)
    Arbitrary conversion from household to individual level in variable insertion; shifts the dollar amounts the LLM sees and hence the advice.
  • Translation prompt heuristics (e.g., target-date fund default allocation) = 90% D + 10% N if age<40; (170-2*age)% D otherwise; 30% D + 70% N if age>70 (Appendix A.5.2)
    Ad hoc rules written by the authors that directly generate asset allocations when the text advice is vague; these rules shape the non-diversified and equity share outputs.
assumptions (6)
  • domain assumption The life cycle model's income process, transition probabilities, mortality, and tax rules are calibrated to SIPP, SSA, and 2025 tax law (Section B.6)
    The benchmark and the simulation environment inherit any errors in these external calibrations.
  • domain assumption Asset returns are uncorrelated with labor income shocks and employment transitions (Section B.6)
    Standard in life cycle models but not tested here; with correlation, optimal equity shares and the LLM comparison would shift.
  • domain assumption The non-diversified assets are calibrated so neither can improve the Sharpe ratio of bond plus diversified index, so the normative model holds zero non-diversified shares (Section B.6)
    Used for the normative benchmark only; the LLM simulation allows all four assets.
  • domain assumption Each LLM query is independent with no memory; the only link across periods is the state evolution (Section 1.3)
    Real ChatGPT use includes memory and follow-up questions, which the simulation abstracts from; the paper acknowledges this in Section 4.2.
  • ad hoc to paper Variable insertion preserves the respondent's writing style and concerns while replacing state-variable numbers (Appendix A.1)
    The mapping table in Table A1 is hand-built; it assumes a prompt written by someone else in the same employment-age-income bucket, after number replacement, represents what the simulated individual would write.
  • ad hoc to paper The dictionary-based topic categories and key words in Table C1 are sufficient to measure prompt and advice content
    The 27 categories are manually defined; regression coefficients in Figure 6 inherit any miscategorization.

how reviews work

0 comments
Cite this review

Pith. "Pith review of AI Financial Advice: Supply, Demand, and Life Cycle Implications." pith.science (2026). https://pith.science/paper/VTOCQLWT

@misc{pith2026260801607,
  author       = {Pith},
  title        = {Pith review of: AI Financial Advice: Supply, Demand, and Life Cycle Implications},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/VTOCQLWT}},
  note         = {Machine review of arXiv:2608.01607}
}
read the original abstract

We ask a representative sample to write prompts seeking spending and investing advice from LLMs, then simulate the lifetime effects of following the advice under realistic asset and labor market conditions. Applying this method to GPT-5.2, we find following the advice would move respondents toward life cycle theory: broader participation in diversified equity funds, age-declining equity shares, and larger savings buffers. Recommendations vary systematically by gender, prior AI experience, and financial literacy. For gender, two-thirds of recommended equity-share differences arise from men and women writing different prompts (demand), while one-third arise from gender labels attached to otherwise identical prompts (supply).

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

65 extracted references · 64 canonical work pages

  1. [1]

    Abdel Haq, Omar, Amitabh Chandra, Tom \'a s Jagelka, Erzo F. P. Luttmer, and Joshua Schwartzstein (2026), Revealing life preferences through LLM s. Working Paper 35185, National Bureau of Economic Research. haq2026revealing

    work page 2026

  2. [2]

    American Economic Review, 93, 193--215

    Agnew, Julie, Pierluigi Balduzzi, and Annika Sund \'e n (2003), Portfolio choice and trading in a large 401(k) plan. American Economic Review, 93, 193--215. Agnew2003

    work page 2003

  3. [3]

    Campbell, Kasper Meisner Nielsen, and Tarun Ramadorai (2020), Sources of inaction in household finance: Evidence from the Danish mortgage market

    Andersen, Steffen, John Y. Campbell, Kasper Meisner Nielsen, and Tarun Ramadorai (2020), Sources of inaction in household finance: Evidence from the Danish mortgage market. American Economic Review, 110, 3184--3230. andersen2020sources

    work page 2020

  4. [4]

    The Review of Financial Studies

    Andries, Marianne, Maxime Bonelli, and David Sraer (2025), Financial advisors and investors' bias. The Review of Financial Studies. Advance article, published October 27, 2025. Andreis2025

    work page 2025

  5. [5]

    Goldstein (2007), Portfolio choice over the life-cycle when the stock and labor markets are cointegrated

    Benzoni, Luca, Pierre Collin-Dufresne , and Robert S. Goldstein (2007), Portfolio choice over the life-cycle when the stock and labor markets are cointegrated. The Journal of Finance, 62, 2123--2167. Benzoni2007

    work page 2007

  6. [6]

    Journal of Public Economics, 88, 1893--1915

    Bergstresser, Daniel and James Poterba (2004), Asset allocation and asset location: Household evidence from the survey of consumer finances. Journal of Public Economics, 88, 1893--1915. bergstresser2004asset

    work page 2004

  7. [7]

    Choi, David Laibson, and Brigitte C

    Beshears, John, James J. Choi, David Laibson, and Brigitte C. Madrian (2018), Behavioral Household Finance . Handbook of Behavioral Economics, 1, 177--276. Beshears2018a

    work page 2018

  8. [8]

    Bessembinder2018

    Bessembinder, Hendrik (2018), Do stocks outperform Treasury bills? Journal of Financial Economics, 129, 440--457. Bessembinder2018

    work page 2018

Show all 65 references
  1. [9]

    bhattacharya2024women

    Bhattacharya, Utpal, Amit Kumar, Sujata Visaria, and Jing Zhao (2024), Do women receive worse financial advice? The Journal of Finance, 79, 3261--3307. bhattacharya2024women

    2024

  2. [10]

    Jin (2026), Behavioral economics of AI : LLM biases and corrections

    Bini, Pietro, Lin William Cong, Xing Huang, and Lawrence J. Jin (2026), Behavioral economics of AI : LLM biases and corrections. Working Paper 34745, National Bureau of Economic Research. Bini2025

    2026

  3. [11]

    Carter, Kyle Herkenhoff, Jonathan Rothbaum, and Lawrence D

    Braxton, J. Carter, Kyle Herkenhoff, Jonathan Rothbaum, and Lawrence D. W. Schmidt (2025), Changing income risk across the US skill distribution: Evidence from a generalized Kalman filter. American Economic Review, 115, 4438--4475. Braxton2025

    2025

  4. [12]

    American Economic Review, 115, 4218--4252

    Bucher-Koenen, Tabea, Andreas Hackethal, Johannes Koenen, and Christine Laudenbach (2025), Gender differences in financial advice. American Economic Review, 115, 4218--4252. bucherkoenen2025gender

    2025

  5. [13]

    Campbell, and Paolo Sodini (2009), Fight or flight? Portfolio rebalancing by individual investors

    Calvet, Laurent E., John Y. Campbell, and Paolo Sodini (2009), Fight or flight? Portfolio rebalancing by individual investors. The Quarterly Journal of Economics, 124, 301--348. Calvet2009

    2009

  6. [14]

    Cocco (2021), Structuring Mortgages for Macroeconomic Stability

    Campbell, John Y., Nuno Clara, and Jo \ a o F. Cocco (2021), Structuring Mortgages for Macroeconomic Stability . The Journal of Finance, 76, 2525--2576. Campbell2021

    2021

  7. [15]

    and Tarun Ramadorai (2026), Household finance in retrospect and prospect

    Campbell, John Y. and Tarun Ramadorai (2026), Household finance in retrospect and prospect. Working Paper 34621, National Bureau of Economic Research. Journal of Finance: Insights and Perspectives, forthcoming. campbell2025household

    2026

  8. [16]

    The Review of Financial Studies, 35, 4016--4054

    Catherine, Sylvain (2022), Countercyclical Labor Income Risk and Portfolio Choices over the Life-Cycle . The Review of Financial Studies, 35, 4016--4054. Catherine2020

    2022

  9. [17]

    (2022), Popular personal financial advice versus the professors

    Choi, James J. (2022), Popular personal financial advice versus the professors. Journal of Economic Perspectives, 36, 167--192. choi2022popular

    2022

  10. [18]

    Madrian (2011), \ 100 bills on the sidewalk: Suboptimal investment in 401(k) plans

    Choi, James J., David Laibson, and Brigitte C. Madrian (2011), \ 100 bills on the sidewalk: Suboptimal investment in 401(k) plans. Review of Economics and Statistics, 93, 748--763. choi2011100

    2011

  11. [19]

    American Economic Review, 115, 3749--3787

    Choukhmane, Taha (2025), Default options and retirement saving dynamics. American Economic Review, 115, 3749--3787. choukhmane2025default

    2025

  12. [20]

    The Journal of Finance, 81, 5--48

    Choukhmane, Taha and Tim de Silva (2026), What drives investors' portfolio choices? separating risk preferences from frictions. The Journal of Finance, 81, 5--48. Choukhmane2026

    2026

  13. [21]

    Choukhmane, Taha, Lucas Goodman, and Cormac O'Dea (2025), Efficiency in household decision-making: Evidence from the retirement savings of U.S. couples. American Economic Review, 115, 1485--1519. choukhmane2025efficiency

    2025

  14. [22]

    Gomes, and Pascal J

    Cocco, Jo \ a o F., Francisco J. Gomes, and Pascal J. Maenhout (2005), Consumption and portfolio choice over the life cycle. The Review of Financial Studies, 18, 491--533. Cocco2005

    2005

  15. [23]

    Palmer (2026), What do LLM s want? Finance and Economics Discussion Series 2026-006, Board of Governors of the Federal Reserve System

    Cook, Thomas R., Sophia Kazinnik, Zach Modig, and Nathan M. Palmer (2026), What do LLM s want? Finance and Economics Discussion Series 2026-006, Board of Governors of the Federal Reserve System. Cook2026

    2026

  16. [24]

    Hubbard, and Daniel T

    Cooley, Philip L., Carl M. Hubbard, and Daniel T. Walz (1998), Retirement savings: Choosing a withdrawal rate that is sustainable. AAII Journal, 20, 16--21. Available via AAII. Cooley1998

    1998

  17. [25]

    Journal of Economic Literature, 47, 448--474

    Croson, Rachel and Uri Gneezy (2009), Gender differences in preferences. Journal of Economic Literature, 47, 448--474. croson2009gender

    2009

  18. [26]

    Rossi (2019), The Promises and Pitfalls of Robo-Advising

    D'Acunto, Francesco, Nagpurnanand Prabhala, and Alberto G. Rossi (2019), The Promises and Pitfalls of Robo-Advising . The Review of Financial Studies, 32, 1983--2020. DAcunto2019

    2019

  19. [27]

    The Quarterly Journal of Economics, 140, 2851--2905

    de Silva, Tim (2025), Insurance versus moral hazard in income-contingent student loan repayment. The Quarterly Journal of Economics, 140, 2851--2905. de2025insurance

    2025

  20. [28]

    Goodman, and Jonathan A

    Duarte, Victor, Julia Fonseca, Aaron S. Goodman, and Jonathan A. Parker (2025), Simple allocation rules and optimal portfolio choice over the lifecycle. Working Paper 29559, National Bureau of Economic Research. Originally issued December 2021. duarte2024simple

    2025

  21. [29]

    Journal of Political Economy, 127, 233--295

    Egan, Mark, Gregor Matvos, and Amit Seru (2019), The market for financial adviser misconduct. Journal of Political Economy, 127, 233--295. Egan2019a

    2019

  22. [30]

    Journal of Financial Economics

    Fedyk, Anastassia, Ali Kakhbod, Peiyao Li, and Ulrike Malmendier (2026), AI and perception biases in investments: An experimental study. Journal of Financial Economics. Accepted. Fedyk2026

    2026

  23. [31]

    Streich (2025), Using large language models for financial advice

    Fieberg, Christian, Lars Hornuf, Maximilian Meiler, and David J. Streich (2025), Using large language models for financial advice. CESifo Working Paper 11666, CESifo. Fieberg2025

    2025

  24. [32]

    Management Science, 62, 3138--3160

    Filippin, Antonio and Paolo Crosetto (2016), A reconsideration of gender differences in risk attitudes. Management Science, 62, 3138--3160. filippin2016reconsideration

    2016

  25. [33]

    Discussion Paper 21323, Centre for Economic Policy Research

    Foltyn, Richard and Jonna Olsson (2026), The worth of a ``wo'': Gender bias in financial advice from LLM s. Discussion Paper 21323, Centre for Economic Policy Research. foltyn2026worth

    2026

  26. [34]

    Gallup News

    Gallup (2025), Americans still turn to people for financial advice. Gallup News. Gallup Panel survey of U.S. adults. Saad2025GallupAdvice

    2025

  27. [35]

    American Economic Review, 109, 2383--2424

    Ganong, Peter and Pascal Noel (2019), Consumer spending during unemployment: Positive and normative implications. American Economic Review, 109, 2383--2424. Ganong2019

    2019

  28. [36]

    The Journal of Finance, 60, 869--904

    Gomes, Francisco and Alexander Michaelides (2005), Optimal life-cycle asset allocation: Understanding the empirical evidence. The Journal of Finance, 60, 869--904. GomesMichaelides2005

    2005

  29. [37]

    (2020), Portfolio Choice over the Life Cycle : A Survey

    Gomes, Francisco J. (2020), Portfolio Choice over the Life Cycle : A Survey . Annual Review of Financial Economics, 12, 277--304. Gomes2020

    2020

  30. [38]

    Journal of Economic Literature, 59, 919--1000

    Gomes, Francisco J., Michael Haliassos, and Tarun Ramadorai (2021), Household Finance . Journal of Economic Literature, 59, 919--1000. Gomes2021b

    2021

  31. [39]

    https://blog.google/products-and-platforms/products/gemini/gemini-3-flash/

    Google DeepMind (2025), Gemini 3 Flash . https://blog.google/products-and-platforms/products/gemini/gemini-3-flash/. gemini2025b

    2025

  32. [40]

    Parker (2002), Consumption over the life cycle

    Gourinchas, Pierre-Olivier and Jonathan A. Parker (2002), Consumption over the life cycle. Econometrica, 70, 47--89. Gourinchas2002

    2002

  33. [41]

    Rossi, Stephen P

    Greig, Fiona, Tarun Ramadorai, Alberto G. Rossi, Stephen P. Utkus, and Ansgar Walther (2025), Human financial advice in the age of automation. Working paper, SSRN. Revise and resubmit at the Journal of Finance; SSRN version last revised July 7, 2025. Greig2025a

    2025

  34. [42]

    American Economic Review, 87, 192--205

    Gruber, Jonathan (1997), The Consumption Smoothing Benefits of Unemployment Insurance . American Economic Review, 87, 192--205. Gruber1997

    1997

  35. [43]

    McQuade (2021), Mortgage Design in an Equilibrium Model of the Housing Market

    Guren, Adam M., Arvind Krishnamurthy, and Timothy J. McQuade (2021), Mortgage Design in an Equilibrium Model of the Housing Market . The Journal of Finance, 76, 113--168. Guren2021b

    2021

  36. [44]

    arXiv preprint arXiv:2502.16879

    Hao, Yuzhi and Danyang Xie (2025), A multi- LLM -agent-based framework for economic and public policy analysis. arXiv preprint arXiv:2502.16879. hao2025multi

    2025

  37. [45]

    Manning (2023), Large language models as simulated economic agents: What can we learn from Homo Silicus ? Working Paper 31122, National Bureau of Economic Research

    Horton, John J., Apostolos Filippas, and Benjamin S. Manning (2023), Large language models as simulated economic agents: What can we learn from Homo Silicus ? Working Paper 31122, National Bureau of Economic Research. Revised February 2026. Horton2023

    2023

  38. [46]

    Power (2025), As more U.S

    J.D. Power (2025), As more U.S. consumers struggle with rising prices, many turn to artificial intelligence for financial advice. Banking and Payments Intelligence Report - Press Release. Published 28 August 2025; based on survey of approximately 4,000 U.S. consumers. JDPower2...

    2025

  39. [47]

    Smith, Jr

    Krusell, Per and Anthony A. Smith, Jr. (1998), Income and Wealth Heterogeneity in the Macroeconomy . Journal of Political Economy, 106, 867--896. Krusell1998

    1998

  40. [48]

    Melzer, and Alessandro Previtero (2021), The misguided beliefs of financial advisors

    Linnainmaa, Juhani T., Brian T. Melzer, and Alessandro Previtero (2021), The misguided beliefs of financial advisors. The Journal of Finance, 76, 587--621. linnainmaa2021misguided

    2021

  41. [49]

    Press Release

    Lloyds Banking Group (2025), Over 28 million adults now using AI tools to help manage their money. Press Release. Published 03 November 2025. Lloyds2025_AI_Money

    2025

  42. [50]

    Mitchell (2017), Optimal financial knowledge and wealth inequality

    Lusardi, Annamaria, Pierre Carl Michaud, and Olivia S. Mitchell (2017), Optimal financial knowledge and wealth inequality. Journal of Political Economy, 125, 431--477. Lusardi2017

    2017

  43. [51]

    Mitchell (2023), The importance of financial literacy: Opening a new field

    Lusardi, Annamaria and Olivia S. Mitchell (2023), The importance of financial literacy: Opening a new field. Journal of Economic Perspectives, 37, 137--154. lusardi2023importance

    2023

  44. [52]

    (1969), Lifetime Portfolio Selection under Uncertainty : The Continuous-Time Case

    Merton, Robert C. (1969), Lifetime Portfolio Selection under Uncertainty : The Continuous-Time Case . Review of Economics and Statistics, 51, 247--257. Merton1969

    1969

  45. [53]

    Journal of Chinese Economic and Business Studies, 23, 509--587

    Mo, Hongwei and Shumiao Ouyang (2025), (generative) AI in financial economics. Journal of Chinese Economic and Business Studies, 23, 509--587. Mo2025

    2025

  46. [54]

    Moss, Austin, Jackie Wegner, and Sarah L. C. Zechman (2026), AI meets DIY : The impact of human intervention on AI -assisted investing. SSRN Electronic Journal. Moss2026

    2026

  47. [55]

    NBER Working Paper 17929, National Bureau of Economic Research

    Mullainathan, Sendhil, Markus Noeth, and Antoinette Schoar (2012), The market for financial advice: An audit study. NBER Working Paper 17929, National Bureau of Economic Research. Mullainathan2012a

    2012

  48. [56]

    https://openai.com/index/introducing-gpt-5-2/

    OpenAI (2025), Introducing GPT-5.2 . https://openai.com/index/introducing-gpt-5-2/. OpenAI Blog. OpenAI2025a

    2025

  49. [57]

    https://openai.com/index/gpt-5-6/

    OpenAI (2026), GPT-5.6 : Frontier intelligence that scales with your ambition. https://openai.com/index/gpt-5-6/. OpenAI Blog. OpenAI2026

    2026

  50. [58]

    Working paper, SSRN; last revised December 8, 2025

    Ouyang, Shumiao, Hayong Yun, and Xingjian Zheng (2025), AI as decision-maker: Ethics and risk preferences of LLM s. Working paper, SSRN; last revised December 8, 2025. Ouyang2025

    2025

  51. [59]

    Journal of Financial Economics, 155, 103829

    Reher, Michael and Stanislav Sokolinski (2024), Robo advisors and access to wealth management. Journal of Financial Economics, 155, 103829. Reher2024

    2024

  52. [60]

    Annual Review of Financial Economics, 16, 391--411

    Reuter, Jonathan and Antoinette Schoar (2024), Demand-side and supply-side constraints in the market for financial advice. Annual Review of Financial Economics, 16, 391--411. reuter2024demand

    2024

  53. [61]

    Lo (2024), LLM economicus? mapping the behavioral biases of LLM s via utility theory

    Ross, Jillian, Yoon Kim, and Andrew W. Lo (2024), LLM economicus? mapping the behavioral biases of LLM s via utility theory. arXiv preprint arXiv:2408.02784. Ross2024

    2024 arXiv

  54. [62]

    and Stephen P

    Rossi, Alberto G. and Stephen P. Utkus (2020), Who Benefits from Robo-advising ? Evidence from Machine Learning . SSRN Electronic Journal. Last revised February 1, 2021. Rossi2020

    2020

  55. [63]

    Working paper, SSRN

    Rumpf, Matthias, Michael Haliassos, Tetyana Kosyakova, and Thomas Otter (2026), From humans to algorithms: How financial advice differs across professionals, peers, and LLM s. Working paper, SSRN. rumpf2026humans

    2026

  56. [64]

    takayanagi2025generative

    Takayanagi, Takehiro, Kiyoshi Izumi, Javier Sanz-Cruzado, Richard McCreadie, and Iadh Ounis (2025), Are generative AI agents effective personalized financial advisors? In Proceedings of the 48th International ACM SIGIR Conference on Research and Development in Information Retr...

    2025

  57. [65]

    Douglas (2025), Consumption and savings with large language model agents

    Verstyuk, Sergiy and Michael R. Douglas (2025), Consumption and savings with large language model agents. Working paper, Harvard University; SSRN version posted January 8, 2026 and last revised January 16, 2026. douglas2024consumption

    2025

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.