{"paper_id":"2404.18017","title":"Application of Deep Learning for Factor Timing in Asset Management","authors":["Prabhu Prasad Panda","Maysam Khodayari Gharanchaei","Xilin Chen","Haoshu Lyu"],"paper_date":"2024-04-27","paper_type":"empirical","paper_url":"https://arxiv.org/pdf/2404.18017v1","page_url":"https://thequant.space/flowcharts/application-of-deep-learning-for-factor-timing-in-asset-mana/","scores":{"empirical_rigor":6.5,"math_complexity":4,"quadrant":"Street Traders"},"proposes":{"status":"available","statement":"The paper evaluates regression models for predicting the CMA factor premium and uses them for factor‑timing investment strategies.","quote":"The paper examines the performance of regression models (OLS linear regression, Ridge regression, Random Forest, and Fully-connected Neural Network) on the prediction of CMA (Conservative Minus Aggressive) factor premium and the performance of factor timing investment with them.","summary":"The paper examines the performance of regression models (OLS linear regression, Ridge regression, Random Forest, and Fully-connected Neural Network) on the prediction of CMA (Conservative Minus Aggressive) factor premium and the performance of factor timing investment with them. Out-of-sample…","strategy":null},"assumptions":{"status":"available","items":[{"assumption":"Investors are risk‑averse and maximize a mean‑variance utility.","type":"model","stated":"explicit","quote":"We assume a risk averse (Sharpe, 1964) and (Welch and Goyal, 2008) utility function 𝐸[𝑟] − 𝛾. 𝜎𝑟2 (Shanken and Weinstein, 1990)"},{"assumption":"The portfolio consists of only the CMA factor and a risk‑free asset.","type":"market","stated":"explicit","quote":"The research focuses on a two-asset portfolio between CMA factor and risk-free asset (Shanken, 1982) and (Shanken and Weinstein, 1990)."},{"assumption":"Optimal weight is computed as 1/γ times expected return over variance.","type":"model","stated":"explicit","quote":"the optimal weight (Welch and Goyal, 2008) in the factor that can maximize the utility: 𝜔𝑡∗= 1/𝛾 . 𝐸[𝑟𝑡+1]/𝜎𝑟2 = 1/𝛾 . 𝑓(𝑿𝑡)/𝜎𝑟2"},{"assumption":"Transaction cost is proportional to the change in position (basis points per trade).","type":"execution","stated":"explicit","quote":"In this session, we assume that the transaction cost is proportional to the change in position."},{"assumption":"When modeling quadratic costs, they are equivalent to linear market impact.","type":"execution","stated":"explicit","quote":"If we assume that the transaction cost is quadratic to the change in position, which is equivalent to linear market impact assumption."},{"assumption":"Training uses an expanding window; NN3 is trained on the initial training period only to avoid over‑fitting.","type":"model","stated":"explicit","quote":"NN3 is trained on the training only without expanding window to avoid overfitting (Merton, 1980) and handle computational complexity"}]},"datasets":{"status":"available","items":[{"name":"Fama‑French factor data","provider":"Kenneth French’s data library","kind":"macro","access":"public","period":"January 1963 to December 2022","universe":null,"quote":"We source our factor data from Fama and French (Breeden, 1979), accessible at Kenneth French’s data library (Breeden et al., 1989) (monthly, January 1963 to December 2022)","known_id":"french-library"},{"name":"Welch and Goyal timing variables","provider":"Goyal’s website","kind":"macro","access":"public","period":"January 1871 to December 2022","universe":null,"quote":"our timing variable data from Welch and Goyal (2008), accessible at Goyal’s website (Campbell and Thompson, 2008) (monthly, January 1871 to December 2022)"}],"known":[{"id":"french-library","name":"Kenneth French Data Library","provider":"Kenneth R. French","kind":"market-prices","access":"public","quote":"DATA We source our factor data from Fama and French (Breeden, 1979), accessible at Kenneth French’s data library (Breeden et al., 1989) (monthly, January 1963 to December 2022), and our timing variable data from Welch and Goyal (2008), accessible at Goyal’s website (Campbell and Thompson, 2008) (monthly, January 1871 to December 2022).","mentions":1,"used_in":1}],"sample":{"start":null,"end":null,"market":null,"asset_class":null,"frequency":null},"data_resolution":"weekly-monthly"},"transaction_costs":{"status":"available","modeled":"yes","headline_net_of_costs":null,"detail":"Proportional costs of 10‑50 basis points are applied; quadratic cost of 20‑50 bps is also examined.","quote":"Starting from proportional transaction cost is 20bps, all models will underperform the constant weighting scheme. If the proportional transaction cost is 50bps, all models will underperform the constant weighting scheme.","source":"front_matter","rule":"R3","candidates":{"front_matter":"yes","evidence_card":"yes","evidence_card_quoted":true},"card_detail":"Proportional costs of 10‑50 basis points are applied; quadratic cost of 20‑50 bps is also examined.","card_quote":"Starting from proportional transaction cost is 20bps, all models will underperform the constant weighting scheme. If the proportional transaction cost is 50bps, all models will underperform the constant weighting scheme.","meaning":"costs are part of the reported headline results"},"results":{"status":"available","evaluation_type":"trading-pnl","findings":[{"finding":"More flexible models (NN3 and random forest) achieve higher out‑of‑sample R‑squared in predicting the CMA premium.","kind":"predictive-accuracy","quote":"The outcome shows that more flexible structures like NN3 and random forest can better explain the variance in the factor premium of the unseen period."},{"finding":"Factor‑timing with random forest yields stable out‑of‑sample wealth growth, the highest Sharpe ratio, and lower downside risk than NN3.","kind":"performance","quote":"According to the wealth paths, factor timing with the random forest provides a stably better performance than both linear candidates and the unconditional optimal weighting, and it entails much lower downside risk than NN3. Among all strategies of factor timing, random forest often yields the highest Sharpe Ratio."},{"finding":"NN3’s optimal weights fluctuate heavily, leading to large return erosion when transaction costs are considered.","kind":"performance","quote":"However, the optimal weights defined with NN3 obviously fluctuates much more fiercely than other models, which can lead to much larger erosion if transaction cost is considered."},{"finding":"When proportional transaction costs are low (e.g., 10 bps) monthly rebalancing and optimal‑interval rebalancing perform similarly; at higher costs (50 bps) optimal rebalancing adds extra return (0.13 % for random forest, 0.63 % for NN3).","kind":"performance","quote":"For instance, if there exists a proportional transaction cost of 50bps, the optimal rebalancing interval for the factor timing with random forest and NN3 can lead to an annualized 0.13% and 0.63% of extra return on the latter 60% of the OOS months."},{"finding":"Ridge regression produces weights similar to OLS under Campbell‑Thompson restrictions but incurs higher turnover and transaction costs.","kind":"performance","quote":"ridge (L2‑regulated) linear regression provides very similar wealth path as the OLS linear regression under Campbell and Thompson restrictions. However, factor timing with ridge will entail a slightly larger position changes than OLS, which can lead to larger erosion if transaction cost is considered."}],"reported_performance":null,"out_of_sample":{"method":"unclear","headline_out_of_sample":null}},"independent_evidence":{"status":"none_found","supports":[],"contradicts":[],"qualifies":[],"compared_with":33,"citations":{"cited_by_in_corpus":0,"by_independent_authors":0,"how_cited":{"used_as_baseline":0,"uses_method_or_data":0,"extends":0,"discusses_findings":0,"mentions":0,"unclassified":0},"independent_papers":[],"scope":"Citations between papers in this corpus only; citations from outside it are not counted."}},"implementation":{"status":"available","public":false,"code_url":null,"note":"No repository presented by the authors as their own was found in the paper. Absence here is not proof that no code exists."},"replication":{"status":"none_found","card":null,"note":"No replication of this paper has been run by thequant.space yet."},"extraction":{"text_source":"fulltext","model":"openai/gpt-oss-120b","version":1,"extracted_at":"2026-11-14T18:10:42Z"},"provenance":{"claims":"Everything here is what the authors state in the paper. Nothing was re-run or reproduced by the platform.","quotes":"Assumptions, datasets, cost detail and findings were extracted by a language model. Every quote is verbatim paper text. Where an item has `support`, \"full\" means every figure in it appears in the quoted sentences and \"partial\" means a figure is stated elsewhere in the paper; items with a figure found nowhere in the paper were discarded.","datasets_known":"datasets.known lists well-known vendors and public datasets by a stable id, found by a dictionary over the full text where the sentence says the data is used; it is independent of the model and covers only names in the dictionary.","flags":"transaction_costs.modeled, results.evaluation_type and results.out_of_sample.method are platform classifications (structured-decision models); see https://thequant.space/score-guide/#penalty-flags","comparisons":"supports / contradicts / qualifies are a language model's reading of two papers' quoted findings, restricted to closely related papers that share no author. They are pointers to read, not verified replications, and coverage is partial.","implementation":"code_url is a repository the paper itself presents as its own (rule-scored, ambiguous cases judged by a model), checked for liveness when it was recorded."},"docs":"https://thequant.space/agents/"}