{"paper_id":"2505.19617","title":"Hybrid Models for Financial Forecasting: Combining Econometric, Machine Learning, and Deep Learning Models","authors":["Dominik Stempień","Robert Ślepaczuk"],"paper_date":"2025-05-26","paper_type":"empirical","paper_url":"https://arxiv.org/pdf/2505.19617v1","page_url":"https://thequant.space/flowcharts/hybrid-models-for-financial-forecasting-combining-econometr/","scores":{"empirical_rigor":9,"math_complexity":7,"quadrant":"Holy Grail"},"proposes":{"status":"available","statement":"The paper proposes hybrid model architectures that combine the econometric ARIMA model with either SVM or LSTM, assuming a non‑additive relationship between linear and nonlinear components, to improve forecasting and trading performance.","quote":"The most effective hybrid model architecture was achieved by combining the econometric ARIMA model with either SVM or LSTM, under the assumption of a non-additive relationship between the linear and nonlinear components.","summary":"This research systematically develops and evaluates various hybrid modeling approaches by combining traditional econometric models (ARIMA and ARFIMA models) with machine learning and deep learning techniques (SVM, XGBoost, and LSTM models) to forecast financial time series. The empirical analysis…","strategy":"LSTM‑ARIMA (1) hybrid model"},"assumptions":{"status":"available","items":[{"assumption":"Econometric models assume linear, stationary, and normal distribution properties in the data.","type":"statistical","stated":"explicit","quote":"This group of methods usually adopts assumptions concerning the linear, stationary, and normal distribution properties present in the data (Shah et al. 2019)."},{"assumption":"Hybrid models assume a non‑additive relationship between linear and nonlinear components.","type":"model","stated":"explicit","quote":"under the assumption of a non-additive relationship between the linear and nonlinear components."},{"assumption":"ARIMA requires the underlying series to be stationary or first transformed into a stationary form.","type":"statistical","stated":"explicit","quote":"In order to apply ARIMA methodology, the underlying series is required to be either stationary or to be first transformed into the stationary form (Kobiela et al., 2022)."},{"assumption":"Trading signals are generated only when predicted returns exceed transaction costs.","type":"execution","stated":"explicit","quote":"To generate a buy signal, the predicted value of the next day’s return must exceed the transaction costs."},{"assumption":"Closing price series is transformed into logarithmic returns for model input.","type":"data","stated":"explicit","quote":"To generate input for the predictive models, the closing price series was transformed into the logarithmic returns."},{"assumption":"Transaction costs are modeled as a percentage parameter c in the signal rule.","type":"execution","stated":"explicit","quote":"c – transaction costs (formulated in percentages)"}]},"datasets":{"status":"available","items":[{"name":"S&P 500 daily price data","provider":null,"kind":"market-prices","access":"unknown","period":"2002-01-01 to 2023-12-31","universe":null,"quote":"... The data period for S&P 500 spanned between 1 January 2002 and 31 December 2023."},{"name":"Bitcoin daily price data","provider":null,"kind":"crypto-onchain","access":"unknown","period":"2015-01-01 to 2023-12-31","universe":null,"quote":"For Bitcoin, the data covered daily observations between 1 January 2015 and 31 December 2023."}],"known":[{"id":"yahoo-finance","name":"Yahoo Finance","provider":"Yahoo","kind":"market-prices","access":"public","quote":"5 3 Data 3.1 Data fetching and preprocessing The data used in the study was downloaded using the Python yfinance library.","mentions":1,"used_in":1}],"sample":{"start":"2018-01","end":"2023-12","market":"Bitcoin","asset_class":"crypto","frequency":"daily"},"data_resolution":"daily"},"transaction_costs":{"status":"available","modeled":"yes","headline_net_of_costs":null,"detail":"c is a percentage representing transaction costs; costs are applied when generating trading signals and double‑charged when switching positions.","quote":"... The process of opening a new position or modifying an existing one is performed only if potential profits from the trade surpass the transaction costs.","source":"evidence_card","rule":"R2","candidates":{"front_matter":"unclear","evidence_card":"yes","evidence_card_quoted":true},"card_detail":"c is a percentage representing transaction costs; costs are applied when generating trading signals and double‑charged when switching positions.","card_quote":"... The process of opening a new position or modifying an existing one is performed only if potential profits from the trade surpass the transaction costs.","meaning":"costs are part of the reported headline results"},"results":{"status":"available","evaluation_type":"trading-pnl","findings":[{"finding":"Hybrid ARIMA‑SVM and ARIMA‑LSTM models outperform their individual components and the Buy&Hold benchmark.","kind":"performance","quote":"The obtained findings indicate that the proper construction process of hybrid models plays a crucial role in developing profitable trading strategies, outperforming their individual components and the benchmark Buy&Hold strategy."},{"finding":"For Bitcoin, the LSTM‑ARIMA (1) hybrid achieves IR = 0.50 and SR = 1.06, surpassing both the market and its constituent models.","kind":"performance","quote":"Long‑Short trading signals applied to the predictions of LSTM‑ARIMA (1) resulted in IR and SR metrics at the levels of 0.50 and 1.06, respectively."},{"finding":"Hybrid SVM‑ARIMA (1), SVM‑ARIMA (2) and LSTM‑ARFIMA (1) also outperform the market and their individual models for Bitcoin.","kind":"performance","quote":"The SVM‑ARIMA (1), SVM‑ARIMA (2), and LSTM‑ARFIMA (1) also managed to outperform both the market and their individual underlying models."},{"finding":"XGBoost‑based hybrid approaches do not improve predictive power and underperform relative to other hybrids.","kind":"negative","quote":"These observations are consistent with the findings for the S&P 500 index: the unsuitability of XGBoost‑based approaches and the superior performance of hybrid methodology not assuming an additive relationship between linear and nonlinear components."},{"finding":"For Bitcoin, the ARIMA model yields the most accurate predictions according to RMSE and MAE, while hybridization does not improve accuracy.","kind":"predictive-accuracy","quote":"The most accurate predictions, according to the RMSE and MAE metrics, were generated by the ARIMA model."}],"reported_performance":{"strategy":"LSTM‑ARIMA (1) hybrid model","sharpe":null,"sharpe_benchmark":null,"annual_return_pct":30,"total_return_pct":null,"max_drawdown_pct":0.61},"out_of_sample":{"method":"unclear","headline_out_of_sample":null}},"independent_evidence":{"status":"none_found","supports":[],"contradicts":[],"qualifies":[],"compared_with":37,"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":"2026-11-14T17:37:10Z"},"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/"}