{"paper_id":"0b607ef5-9919-483e-be36-e3aeb168fd2d","body_text":"Impact of Hyperparameter Optimisation Techniques in Deep Learning-based Investment Predictions: An Indian ETF-based analysis | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Systematic Review Impact of Hyperparameter Optimisation Techniques in Deep Learning-based Investment Predictions: An Indian ETF-based analysis Alan Vellaiparambill, N Natchimuthu This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7567095/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract The integration of deep learning into financial forecasting has significantly advanced predictive analytics, yet the effectiveness of these models is critically dependent on hyperparameter optimization (HPO). This study investigates the role of HPO in enhancing the predictive and financial performance of Long Short-Term Memory (LSTM) and one-dimensional Convolutional Neural Network (1D-CNN) models applied to the Nifty BeEs Exchange Traded Fund (ETF), a key proxy for the Indian equity market. Using daily log-return data from 2010 to 2025, four HPO techniques grid search, Bayesian optimization, Optuna GridSampler, and Optuna Tree-structured Parzen Estimator (TPE) were systematically compared. Evaluation metrics included Root Mean Squared Error (RMSE), directional accuracy (DA), Sharpe ratios, and computational cost. Results demonstrate that while traditional methods provide modest improvements, they fail to align statistical accuracy with financial viability. In contrast, Optuna-based approaches, particularly the TPE Sampler, significantly improved outcomes, raising LSTM accuracy to 63% and CNN accuracy to 61%, with Sharpe ratios exceeding 1.2 at minimal computational cost. These findings underscore that hyperparameter optimization is not a peripheral technical task but a strategic determinant of investment applicability, transforming deep learning models from theoretical constructs into practical forecasting engines. The study contributes to bridging methodological innovations in computer science with financial econometrics, offering actionable insights for ETF prediction in emerging markets. Deep learning Index Prediction Hyperparameter Optimization Financial Forecasting Figures Figure 1 Figure 6 1. Introduction The integration of artificial intelligence (AI) into financial markets has reshaped predictive analytics, particularly in investment forecasting. Market dynamics are notoriously difficult to predict due to high volatility, nonlinear dependencies, and the constant influence of global macroeconomic factors. Exchange Traded Funds (ETFs), such as India’s Nifty BeEs, provide diversified market exposure and serve as valuable instruments for both institutional and retail investors. Given their strong correlation with the NIFTY 50 index, forecasting ETF behaviour carries substantial implications for portfolio management, trading strategies, and market stability. Deep learning models such as Long Short-Term Memory (LSTM) networks and one-dimensional Convolutional Neural Networks (1D-CNNs) have gained traction in financial forecasting because of their ability to model temporal and sequential data. While LSTMs excel at capturing long-range dependencies, CNNs effectively extract local temporal features, making them suitable complements in market prediction tasks. However, their predictive capacity and reliability depend heavily on hyperparameter optimization (HPO). Hyperparameters including learning rate, batch size, dropout ratio, and hidden unit configuration directly affect convergence, generalization, and stability. Poorly optimized models risk overfitting or underperformance, even if the underlying architecture is theoretically sound. Traditional optimization methods such as grid search and Bayesian optimization remain widely used in finance but face limitations: grid search is exhaustive yet computationally expensive, while Bayesian optimization can converge prematurely and is sensitive to prior assumptions. Recent advancements in HPO frameworks, particularly Optuna, offer more efficient alternatives. Optuna’s GridSampler introduces structured yet efficient search, while its Tree-structured Parzen Estimator (TPE) Sampler employs probabilistic strategies that adaptively balance exploration and exploitation. Such methods reduce computational costs while improving predictive robustness, making them highly relevant for financial applications. This study evaluates the impact of HPO on deep learning-based investment prediction, using daily log-return data from Nifty BeEs between 2010 and 2025. By comparing four optimization strategies grid search, Bayesian optimization, Optuna GridSampler, and Optuna TPE Sampler the research investigates not only predictive accuracy but also financial performance through Sharpe ratios. Results indicate that Optuna-based approaches, particularly TPE, significantly outperform traditional methods, demonstrating superior accuracy, profitability, and efficiency. The study contributes to financial econometrics by bridging methodological innovation with applied investment analytics, offering insights for both academia and practice. 2. Literature Review Financial econometrics has long sought to improve predictive modeling in markets characterized by volatility clustering, fat-tailed distributions, and non-stationarity. Traditional econometric approaches, including autoregressive integrated moving average (ARIMA) and generalized autoregressive conditional heteroskedasticity (GARCH), provided foundational insights but struggled with nonlinear and chaotic price dynamics. This limitation led to the adoption of machine learning and, more recently, deep learning, which offer greater flexibility in capturing complex dependencies. Neural architectures such as RNNs and LSTMs have been applied extensively to financial forecasting. Levchenko et al. [ 1 ] demonstrated that chain-structured neural architecture search improves both accuracy and stability in financial time series, highlighting the value of adaptive model design. Kehinde et al. [ 2 ] extended this innovation with Helformer, an attention-based model tailored for cryptocurrency prediction, showing how hybrid architectures can accommodate extreme volatility. These studies underscore the adaptability of deep learning in financial contexts but also point to the need for robust optimization to maximize performance. Parallel research has emphasized optimization as a decisive factor in model effectiveness. Gonzalvez et al. [ 3 ] highlighted the potential of Gaussian processes and Bayesian optimization for refining financial models, though such methods remain computationally demanding. Bayesian optimization is particularly valuable for navigating non-convex landscapes but risks being trapped in local optima, especially with high-dimensional parameter spaces. The emergence of Optuna marked a significant methodological leap. Its GridSampler balances structured exploration with computational feasibility, while the TPE Sampler employs adaptive probabilistic modeling that dynamically focuses on promising regions of the search space [ 4 – 6 ]. These innovations have been shown to reduce training time while improving accuracy, though their application in emerging financial markets remains limited. Critically, scholars have argued that statistical accuracy does not automatically translate into financial relevance. Zhang et al. [ 7 ] stressed the importance of directional accuracy and profitability over traditional error metrics. Li and Hoi [ 8 ] similarly emphasized aligning optimization outcomes with financial utility measures such as the Sharpe ratio. In the Indian context, while studies have investigated AI in equities and derivatives, applications to ETFs remain sparse, representing an underexplored but highly relevant domain [ 9 , 10 ]. Taken together, the literature suggests two core insights: model architecture alone cannot ensure predictive success without robust HPO, and adaptive frameworks like Optuna may provide the best balance of accuracy, efficiency, and financial viability. These insights form the conceptual foundation for the present study. 3. Research Objectives This study aims to evaluate the role of hyperparameter optimization in shaping the predictive accuracy and financial viability of deep learning models applied to the Indian ETF market. While LSTMs and CNNs are powerful for modeling sequential dependencies, their effectiveness in investment contexts depends critically on fine-tuning hyperparameters. The first objective is to compare four HPO strategies grid search, Bayesian optimization, Optuna GridSampler, and Optuna TPE Sampler using daily log-return data from Nifty BeEs (2010–2025). Predictive performance is measured by Root Mean Squared Error (RMSE) and directional accuracy. The second objective is to evaluate financial performance. By analyzing Sharpe ratios derived from trading strategies based on predictions, the study links statistical improvements to risk-adjusted profitability. The third objective is to assess computational efficiency. While accuracy is valuable, practical deployment requires methods that converge quickly with limited resource consumption. Comparing computational costs across HPO techniques ensures relevance to real-world applications. Ultimately, the research seeks to bridge methodological innovation and financial applicability, providing insights into how HPO transforms deep learning models from academic prototypes into practical trading tools. 4. Research Methodology The study focuses on Nifty BeEs, India’s most liquid ETF and a close proxy for the NIFTY 50 index. Historical adjusted daily closing prices were obtained from Yahoo Finance, covering January 2010 to February 2025. Log returns, defined as \\(\\:{r}_{t}=\\text{ln}\\left({P}_{t}\\right)-\\text{l}\\text{n}\\left({P}_{t-1}\\right)\\) , were computed to ensure stationarity and stabilize variance. Data preprocessing included removal of duplicates, handling of missing values, and MinMax normalization of inputs to [0,1]. Sequences of 60-day windows were constructed to predict the subsequent daily return. The dataset was split into training (70%), validation (15%), and test (15%) subsets. Two models were implemented: LSTM: Designed to capture long-range dependencies. 1D-CNN: Efficient at extracting local temporal patterns. Both models were implemented in TensorFlow/Keras, maintaining identical input-output structures for comparability.Hyperparameters tuned included hidden units, dropout rates, learning rates, batch sizes, and epochs. Four optimization methods were employed: Grid Search: Exhaustive but computationally intensive. Bayesian Optimization: Probabilistic refinement with moderate efficiency. Optuna GridSampler: Adaptive structured search. Optuna TPE Sampler: Probabilistic adaptive strategy. Evaluation metrics included RMSE, directional accuracy (proportion of correctly predicted price movements), Sharpe ratios (measuring risk-adjusted returns), and computational cost (runtime and resource use). 5. Results The choice of HPO method significantly influenced both predictive accuracy and financial outcomes. Baseline LSTM models optimized with grid search yielded RMSE of 0.0068, directional accuracy of 54.2%, and Sharpe ratios of 0.70. Bayesian optimization provided marginal gains, raising accuracy to 57–59% but Sharpe ratios remained below 1.0, indicating weak financial viability. Optuna-based methods marked a decisive improvement. The GridSampler achieved 60% accuracy for both LSTM and CNN models, producing Sharpe ratios above 1.0 with lower computational cost than traditional methods. The TPE Sampler delivered the strongest results, raising LSTM accuracy to 63%, CNN accuracy to 61%, and Sharpe ratios above 1.2, all at minimal resource cost. Table 1 HPO Technique and its impact Method DA (LSTM) DA (CNN) Sharpe Ratio Computational Cost Grid Search 58% 56% Low High Bayesian Optimisation 59% 57% Low-Moderate Moderate Optuna GridSampler 60% 60% Moderate Low Optuna TPESampler 63% 61% High Low The comparative results confirm that adaptive optimization not only improves predictive precision but also enhances portfolio profitability. Computational efficiency further underscores their practical relevance, with Optuna methods converging faster than traditional approaches. 6. Discussion The findings underscore three key insights. First, statistical accuracy alone does not guarantee financial utility. Both grid search and Bayesian optimization improved error metrics but failed to deliver Sharpe ratios above unity, confirming that financial evaluation must go beyond RMSE. Second, Optuna’s adaptive strategies, particularly TPE, demonstrated clear superiority. By dynamically balancing exploration and exploitation, TPE avoided premature convergence and identified hyperparameter configurations that maximized both predictive robustness and profitability. These results are especially relevant in volatile emerging markets such as India, where ETF dynamics are shaped by policy, liquidity, and global capital flows.Third, computational efficiency is not merely a technical concern but a strategic necessity. Models that recalibrate quickly enable near real-time deployment, reducing risk exposure and opportunity costs. Optuna’s ability to combine higher accuracy with faster convergence makes it well-suited for both institutional and retail adoption. The broader implication is that HPO is not a peripheral task but a critical determinant of whether deep learning models succeed in financial applications. Adaptive probabilistic approaches elevate models from experimental tools to viable forecasting engines, bridging the gap between computational innovation and financial decision-making. 7. Conclusion This study evaluated the impact of hyperparameter optimization on deep learning-based ETF prediction, focusing on the Nifty BeEs. By comparing grid search, Bayesian optimization, Optuna GridSampler, and Optuna TPE Sampler, the research demonstrated that optimization methods fundamentally alter predictive accuracy, financial relevance, and efficiency. Traditional methods yielded modest improvements but failed to align predictive performance with investment profitability. In contrast, Optuna-based approaches consistently outperformed. GridSampler delivered balanced gains in accuracy and Sharpe ratios above 1.0, while TPE achieved the highest accuracy (63% for LSTM, 61% for CNN) and Sharpe ratios above 1.2 at minimal computational cost.The findings highlight that hyperparameter optimization is not ancillary but central to deploying deep learning in finance. Adaptive methods enable models to transcend statistical benchmarks and deliver actionable financial value. For Indian ETFs, these results confirm that advanced optimization can transform predictive analytics into a practical tool for risk-adjusted returns. This research contributes to bridging computer science and financial econometrics, offering theoretical and practical insights. Future work could extend this framework to high-frequency data, cross-asset ETFs, or hybrid models incorporating macroeconomic indicators. Ultimately, integrating advanced HPO methods into financial modeling represents a decisive step toward unlocking the full potential of AI in capital markets. Declarations Conflict of interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. Funding No funding is received pertaining to this research. Author Contribution Alan Vellaiparambill wrote the main manuscript text and E.F. prepared figuresDr. Natchimuthu N reviewed the manuscript and supported the HPO process at ideation and implementation Acknowledgements This work was supported by Department of Commerce, Christ Deemed to be University, India. Data Availability The data that support the findings of this study are openly available at the websites linked in the article. References Levchenko O, Lemeshko O, Skorokhod Y, Khudov H, Khudov H (2024) Chain-structured neural architecture search for financial time series forecasting. J Comput Finance Analytics 18(2):101–120 Kehinde O, Li Y, Zhang H (2025) Helformer: Attention-based deep learning for cryptocurrency price prediction. Expert Syst Appl 240:121245 Gonzalvez J, Patel R, Singh A (2019) Bayesian optimization and Gaussian processes in financial modeling. Quant Finance 19(7):1125–1141 Akiba T, Sano S, Yanase T, Ohta T, Koyama M (2019) Optuna: A next-generation hyperparameter optimization framework. Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, 2623–2631 Bergstra J, Bardenet R, Bengio Y, Kégl B (2011) Algorithms for hyper-parameter optimization. Adv Neural Inf Process Syst 24:2546–2554 Bergstra J, Yamins D, Cox D (2013) Making a science of model search: Hyperparameter optimization in hundreds of dimensions for vision architectures. Proceedings of the 30th International Conference on Machine Learning, 115–123 Zhang L, Wu J, Zhou D (2020) Beyond accuracy: Financial utility metrics in evaluating machine learning models. J Financial Data Sci 2(3):55–72 Li X, Hoi SC (2021) Evaluating financial machine learning models with utility-driven metrics. ACM Trans Manage Inform Syst 12(4):1–22 Ramesh A, Nair S (2022) Deep learning adoption in Indian equity markets: Evidence from NIFTY 50. Asian J Finance Acc 14(1):42–61 Gupta V, Iyer P (2023) Machine learning for derivatives pricing and ETF forecasting in India. Int J Financial Stud 11(2):33–49 Additional Declarations No competing interests reported. Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {\"props\":{\"pageProps\":{\"initialData\":{\"identity\":\"rs-7567095\",\"acceptedTermsAndConditions\":true,\"allowDirectSubmit\":true,\"archivedVersions\":[],\"articleType\":\"Systematic Review\",\"associatedPublications\":[],\"authors\":[{\"id\":516776637,\"identity\":\"f058bb8a-b11c-44b5-b5a3-232fdd292f1b\",\"order_by\":0,\"name\":\"Alan Vellaiparambill\",\"email\":\"data:image/png;base64,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\",\"orcid\":\"\",\"institution\":\"Christ Deemed to be University\",\"correspondingAuthor\":true,\"prefix\":\"\",\"firstName\":\"Alan\",\"middleName\":\"\",\"lastName\":\"Vellaiparambill\",\"suffix\":\"\"},{\"id\":516776638,\"identity\":\"cacc7520-04c2-4b4a-af12-767eab1547e7\",\"order_by\":1,\"name\":\"N Natchimuthu\",\"email\":\"\",\"orcid\":\"\",\"institution\":\"Christ Deemed to be University\",\"correspondingAuthor\":false,\"prefix\":\"\",\"firstName\":\"N\",\"middleName\":\"\",\"lastName\":\"Natchimuthu\",\"suffix\":\"\"}],\"badges\":[],\"createdAt\":\"2025-09-08 19:09:14\",\"currentVersionCode\":1,\"declarations\":{\"humanSubjects\":false,\"vertebrateSubjects\":false,\"conflictsOfInterestStatement\":false,\"humanSubjectEthicalGuidelines\":false,\"humanSubjectConsent\":false,\"humanSubjectClinicalTrial\":false,\"humanSubjectCaseReport\":false,\"vertebrateSubjectEthicalGuidelines\":false},\"doi\":\"10.21203/rs.3.rs-7567095/v1\",\"doiUrl\":\"https://doi.org/10.21203/rs.3.rs-7567095/v1\",\"draftVersion\":[],\"editorialEvents\":[],\"editorialNote\":\"\",\"failedWorkflow\":false,\"files\":[{\"id\":95809427,\"identity\":\"8e6de530-6799-4fb1-b816-82ead82bae0f\",\"added_by\":\"auto\",\"created_at\":\"2025-11-13 08:50:26\",\"extension\":\"docx\",\"order_by\":0,\"title\":\"\",\"display\":\"\",\"copyAsset\":false,\"role\":\"acdc-reference\",\"size\":84868,\"visible\":true,\"origin\":\"\",\"legend\":\"\",\"description\":\"\",\"filename\":\"FIGAlanVellaiparambillArticleSpecialIssueMematicComputing.docx\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-7567095/v1/9ecae21f78a9022283905566.docx\"},{\"id\":95809402,\"identity\":\"4ef11741-2a53-44b7-8989-bd4e20478fbf\",\"added_by\":\"auto\",\"created_at\":\"2025-11-13 08:50:23\",\"extension\":\"docx\",\"order_by\":1,\"title\":\"\",\"display\":\"\",\"copyAsset\":false,\"role\":\"acdc-reference\",\"size\":95613,\"visible\":true,\"origin\":\"\",\"legend\":\"\",\"description\":\"\",\"filename\":\"AlanVellaiparambillArticleSpecialIssueMematicComputing.docx\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-7567095/v1/c8256a419f752fb7fc64fb43.docx\"},{\"id\":95809411,\"identity\":\"cc2dbc9b-2587-42d1-9a56-0780a2c52495\",\"added_by\":\"auto\",\"created_at\":\"2025-11-13 08:50:24\",\"extension\":\"docx\",\"order_by\":2,\"title\":\"\",\"display\":\"\",\"copyAsset\":false,\"role\":\"acdc-reference\",\"size\":25215,\"visible\":true,\"origin\":\"\",\"legend\":\"\",\"description\":\"\",\"filename\":\"TableAlanVellaiparambillArticleSpecialIssueMematicComputing.docx\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-7567095/v1/e2ad9c101db66cb9a7d94af4.docx\"},{\"id\":95809439,\"identity\":\"9e6c26c2-c001-48ab-a6c5-12b811b32858\",\"added_by\":\"auto\",\"created_at\":\"2025-11-13 08:50:26\",\"extension\":\"json\",\"order_by\":3,\"title\":\"\",\"display\":\"\",\"copyAsset\":false,\"role\":\"acdc-reference\",\"size\":4313,\"visible\":true,\"origin\":\"\",\"legend\":\"\",\"description\":\"\",\"filename\":\"d63f55959b504e978abb06e312118fdc.json\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-7567095/v1/99df8f4efac0a19269075cfb.json\"},{\"id\":95809500,\"identity\":\"a6fcf378-21e4-428a-9085-54bc811387ec\",\"added_by\":\"auto\",\"created_at\":\"2025-11-13 08:50:27\",\"extension\":\"xml\",\"order_by\":4,\"title\":\"\",\"display\":\"\",\"copyAsset\":false,\"role\":\"acdc-reference\",\"size\":37786,\"visible\":true,\"origin\":\"\",\"legend\":\"\",\"description\":\"\",\"filename\":\"d63f55959b504e978abb06e312118fdc1enriched.xml\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-7567095/v1/76ff58124c892027186b2378.xml\"},{\"id\":95809567,\"identity\":\"4f512f2c-6585-49af-af2f-d701ec6e50c4\",\"added_by\":\"auto\",\"created_at\":\"2025-11-13 08:50:31\",\"extension\":\"jpeg\",\"order_by\":7,\"title\":\"\",\"display\":\"\",\"copyAsset\":false,\"role\":\"acdc-reference\",\"size\":179169,\"visible\":true,\"origin\":\"\",\"legend\":\"\",\"description\":\"\",\"filename\":\"floatimage3.jpeg\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-7567095/v1/9bf69db399246b359d3e53ef.jpeg\"},{\"id\":95809497,\"identity\":\"f8a28332-5e86-4a35-993d-6a37ce4df8c1\",\"added_by\":\"auto\",\"created_at\":\"2025-11-13 08:50:27\",\"extension\":\"png\",\"order_by\":8,\"title\":\"\",\"display\":\"\",\"copyAsset\":false,\"role\":\"acdc-reference\",\"size\":28808,\"visible\":true,\"origin\":\"\",\"legend\":\"\",\"description\":\"\",\"filename\":\"Onlinefloatimage1.png\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-7567095/v1/5913d5e7e5f144d590771bc3.png\"},{\"id\":95809656,\"identity\":\"edea578f-ca09-44c9-a02d-db7dcf8a40a1\",\"added_by\":\"auto\",\"created_at\":\"2025-11-13 08:50:36\",\"extension\":\"png\",\"order_by\":9,\"title\":\"\",\"display\":\"\",\"copyAsset\":false,\"role\":\"acdc-reference\",\"size\":28808,\"visible\":true,\"origin\":\"\",\"legend\":\"\",\"description\":\"\",\"filename\":\"Onlinefloatimage1.png\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-7567095/v1/79e3aa3849f5032a5bd635d9.png\"},{\"id\":95809548,\"identity\":\"1318a2ab-3f0d-41c8-b022-4b6eb4cc9d42\",\"added_by\":\"auto\",\"created_at\":\"2025-11-13 08:50:31\",\"extension\":\"png\",\"order_by\":10,\"title\":\"\",\"display\":\"\",\"copyAsset\":false,\"role\":\"acdc-reference\",\"size\":45673,\"visible\":true,\"origin\":\"\",\"legend\":\"\",\"description\":\"\",\"filename\":\"Onlinefloatimage3.png\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-7567095/v1/d6ff8437fcb98babd182e250.png\"},{\"id\":95809494,\"identity\":\"db53f8b6-0f73-4651-a137-c421c44bfa31\",\"added_by\":\"auto\",\"created_at\":\"2025-11-13 08:50:27\",\"extension\":\"xml\",\"order_by\":11,\"title\":\"\",\"display\":\"\",\"copyAsset\":false,\"role\":\"acdc-reference\",\"size\":36665,\"visible\":true,\"origin\":\"\",\"legend\":\"\",\"description\":\"\",\"filename\":\"d63f55959b504e978abb06e312118fdc1structuring.xml\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-7567095/v1/543d2a0bd7789d9807c69822.xml\"},{\"id\":95809579,\"identity\":\"18c39cac-7334-46a1-92b7-8e3d0bad5023\",\"added_by\":\"auto\",\"created_at\":\"2025-11-13 08:50:32\",\"extension\":\"html\",\"order_by\":12,\"title\":\"\",\"display\":\"\",\"copyAsset\":false,\"role\":\"acdc-reference\",\"size\":41225,\"visible\":true,\"origin\":\"\",\"legend\":\"\",\"description\":\"\",\"filename\":\"earlyproof.html\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-7567095/v1/f9846ac6ca08f3d53086d92f.html\"},{\"id\":95809503,\"identity\":\"7249a1c6-c698-41d8-9a61-d173bed076ed\",\"added_by\":\"auto\",\"created_at\":\"2025-11-13 08:50:27\",\"extension\":\"jpeg\",\"order_by\":1,\"title\":\"Figure 1\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":102954,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003eCorrelation of Index and Index ETF.\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"floatimage1.jpeg\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-7567095/v1/9075094118d1d1f89ef876ce.jpeg\"},{\"id\":95809395,\"identity\":\"84578722-57db-4b7c-97f7-5412b63de0b2\",\"added_by\":\"auto\",\"created_at\":\"2025-11-13 08:50:22\",\"extension\":\"jpeg\",\"order_by\":6,\"title\":\"Figure 6\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":102954,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003eCorrelation of Index and Index ETF.\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"floatimage1.jpeg\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-7567095/v1/417331d42bb35fbae74b22a8.jpeg\"},{\"id\":96605295,\"identity\":\"86124733-5d5b-49c0-8db8-3bdb3e50e13b\",\"added_by\":\"auto\",\"created_at\":\"2025-11-24 09:22:07\",\"extension\":\"pdf\",\"order_by\":0,\"title\":\"\",\"display\":\"\",\"copyAsset\":false,\"role\":\"manuscript-pdf\",\"size\":568114,\"visible\":true,\"origin\":\"\",\"legend\":\"\",\"description\":\"\",\"filename\":\"manuscript.pdf\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-7567095/v1/829fc023-84c4-4f2a-a018-a129b6d3d841.pdf\"}],\"financialInterests\":\"No competing interests reported.\",\"formattedTitle\":\"Impact of Hyperparameter Optimisation Techniques in Deep Learning-based Investment Predictions: An Indian ETF-based analysis\",\"fulltext\":[{\"header\":\"1. Introduction\",\"content\":\"\\u003cp\\u003eThe integration of artificial intelligence (AI) into financial markets has reshaped predictive analytics, particularly in investment forecasting. Market dynamics are notoriously difficult to predict due to high volatility, nonlinear dependencies, and the constant influence of global macroeconomic factors. Exchange Traded Funds (ETFs), such as India\\u0026rsquo;s Nifty BeEs, provide diversified market exposure and serve as valuable instruments for both institutional and retail investors. Given their strong correlation with the NIFTY 50 index, forecasting ETF behaviour carries substantial implications for portfolio management, trading strategies, and market stability.\\u003c/p\\u003e\\u003cp\\u003eDeep learning models such as Long Short-Term Memory (LSTM) networks and one-dimensional Convolutional Neural Networks (1D-CNNs) have gained traction in financial forecasting because of their ability to model temporal and sequential data. While LSTMs excel at capturing long-range dependencies, CNNs effectively extract local temporal features, making them suitable complements in market prediction tasks. However, their predictive capacity and reliability depend heavily on hyperparameter optimization (HPO). Hyperparameters including learning rate, batch size, dropout ratio, and hidden unit configuration directly affect convergence, generalization, and stability. Poorly optimized models risk overfitting or underperformance, even if the underlying architecture is theoretically sound. Traditional optimization methods such as grid search and Bayesian optimization remain widely used in finance but face limitations: grid search is exhaustive yet computationally expensive, while Bayesian optimization can converge prematurely and is sensitive to prior assumptions. Recent advancements in HPO frameworks, particularly Optuna, offer more efficient alternatives. Optuna\\u0026rsquo;s GridSampler introduces structured yet efficient search, while its Tree-structured Parzen Estimator (TPE) Sampler employs probabilistic strategies that adaptively balance exploration and exploitation. Such methods reduce computational costs while improving predictive robustness, making them highly relevant for financial applications.\\u003c/p\\u003e\\u003cp\\u003eThis study evaluates the impact of HPO on deep learning-based investment prediction, using daily log-return data from Nifty BeEs between 2010 and 2025. By comparing four optimization strategies grid search, Bayesian optimization, Optuna GridSampler, and Optuna TPE Sampler the research investigates not only predictive accuracy but also financial performance through Sharpe ratios. Results indicate that Optuna-based approaches, particularly TPE, significantly outperform traditional methods, demonstrating superior accuracy, profitability, and efficiency. The study contributes to financial econometrics by bridging methodological innovation with applied investment analytics, offering insights for both academia and practice.\\u003c/p\\u003e\"},{\"header\":\"2. Literature Review\",\"content\":\"\\u003cp\\u003eFinancial econometrics has long sought to improve predictive modeling in markets characterized by volatility clustering, fat-tailed distributions, and non-stationarity. Traditional econometric approaches, including autoregressive integrated moving average (ARIMA) and generalized autoregressive conditional heteroskedasticity (GARCH), provided foundational insights but struggled with nonlinear and chaotic price dynamics. This limitation led to the adoption of machine learning and, more recently, deep learning, which offer greater flexibility in capturing complex dependencies.\\u003c/p\\u003e\\u003cp\\u003eNeural architectures such as RNNs and LSTMs have been applied extensively to financial forecasting. Levchenko et al. [\\u003cspan citationid=\\\"CR1\\\" class=\\\"CitationRef\\\"\\u003e1\\u003c/span\\u003e] demonstrated that chain-structured neural architecture search improves both accuracy and stability in financial time series, highlighting the value of adaptive model design. Kehinde et al. [\\u003cspan citationid=\\\"CR2\\\" class=\\\"CitationRef\\\"\\u003e2\\u003c/span\\u003e] extended this innovation with Helformer, an attention-based model tailored for cryptocurrency prediction, showing how hybrid architectures can accommodate extreme volatility. These studies underscore the adaptability of deep learning in financial contexts but also point to the need for robust optimization to maximize performance. Parallel research has emphasized optimization as a decisive factor in model effectiveness. Gonzalvez et al. [\\u003cspan citationid=\\\"CR3\\\" class=\\\"CitationRef\\\"\\u003e3\\u003c/span\\u003e] highlighted the potential of Gaussian processes and Bayesian optimization for refining financial models, though such methods remain computationally demanding. Bayesian optimization is particularly valuable for navigating non-convex landscapes but risks being trapped in local optima, especially with high-dimensional parameter spaces.\\u003c/p\\u003e\\u003cp\\u003eThe emergence of Optuna marked a significant methodological leap. Its GridSampler balances structured exploration with computational feasibility, while the TPE Sampler employs adaptive probabilistic modeling that dynamically focuses on promising regions of the search space [\\u003cspan additionalcitationids=\\\"CR5\\\" citationid=\\\"CR4\\\" class=\\\"CitationRef\\\"\\u003e4\\u003c/span\\u003e\\u0026ndash;\\u003cspan citationid=\\\"CR6\\\" class=\\\"CitationRef\\\"\\u003e6\\u003c/span\\u003e]. These innovations have been shown to reduce training time while improving accuracy, though their application in emerging financial markets remains limited. Critically, scholars have argued that statistical accuracy does not automatically translate into financial relevance. Zhang et al. [\\u003cspan citationid=\\\"CR7\\\" class=\\\"CitationRef\\\"\\u003e7\\u003c/span\\u003e] stressed the importance of directional accuracy and profitability over traditional error metrics. Li and Hoi [\\u003cspan citationid=\\\"CR8\\\" class=\\\"CitationRef\\\"\\u003e8\\u003c/span\\u003e] similarly emphasized aligning optimization outcomes with financial utility measures such as the Sharpe ratio. In the Indian context, while studies have investigated AI in equities and derivatives, applications to ETFs remain sparse, representing an underexplored but highly relevant domain [\\u003cspan citationid=\\\"CR9\\\" class=\\\"CitationRef\\\"\\u003e9\\u003c/span\\u003e, \\u003cspan citationid=\\\"CR10\\\" class=\\\"CitationRef\\\"\\u003e10\\u003c/span\\u003e]. Taken together, the literature suggests two core insights: model architecture alone cannot ensure predictive success without robust HPO, and adaptive frameworks like Optuna may provide the best balance of accuracy, efficiency, and financial viability. These insights form the conceptual foundation for the present study.\\u003c/p\\u003e\"},{\"header\":\"3. Research Objectives\",\"content\":\"\\u003cp\\u003eThis study aims to evaluate the role of hyperparameter optimization in shaping the predictive accuracy and financial viability of deep learning models applied to the Indian ETF market. While LSTMs and CNNs are powerful for modeling sequential dependencies, their effectiveness in investment contexts depends critically on fine-tuning hyperparameters.\\u003c/p\\u003e\\u003cp\\u003eThe first objective is to compare four HPO strategies grid search, Bayesian optimization, Optuna GridSampler, and Optuna TPE Sampler using daily log-return data from Nifty BeEs (2010\\u0026ndash;2025). Predictive performance is measured by Root Mean Squared Error (RMSE) and directional accuracy. The second objective is to evaluate financial performance. By analyzing Sharpe ratios derived from trading strategies based on predictions, the study links statistical improvements to risk-adjusted profitability. The third objective is to assess computational efficiency. While accuracy is valuable, practical deployment requires methods that converge quickly with limited resource consumption. Comparing computational costs across HPO techniques ensures relevance to real-world applications. Ultimately, the research seeks to bridge methodological innovation and financial applicability, providing insights into how HPO transforms deep learning models from academic prototypes into practical trading tools.\\u003c/p\\u003e\"},{\"header\":\"4. Research Methodology\",\"content\":\"\\u003cp\\u003eThe study focuses on Nifty BeEs, India\\u0026rsquo;s most liquid ETF and a close proxy for the NIFTY 50 index. Historical adjusted daily closing prices were obtained from Yahoo Finance, covering January 2010 to February 2025. Log returns, defined as \\u003cspan class=\\\"InlineEquation\\\"\\u003e\\u003cspan class=\\\"mathinline\\\"\\u003e\\\\(\\\\:{r}_{t}=\\\\text{ln}\\\\left({P}_{t}\\\\right)-\\\\text{l}\\\\text{n}\\\\left({P}_{t-1}\\\\right)\\\\)\\u003c/span\\u003e\\u003c/span\\u003e, were computed to ensure stationarity and stabilize variance.\\u003c/p\\u003e\\u003cp\\u003eData preprocessing included removal of duplicates, handling of missing values, and MinMax normalization of inputs to [0,1]. Sequences of 60-day windows were constructed to predict the subsequent daily return. The dataset was split into training (70%), validation (15%), and test (15%) subsets.\\u003c/p\\u003e\\u003cp\\u003eTwo models were implemented:\\u003c/p\\u003e\\u003cp\\u003e\\u003cul\\u003e\\u003cli\\u003e\\u003cp\\u003eLSTM: Designed to capture long-range dependencies.\\u003c/p\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cp\\u003e1D-CNN: Efficient at extracting local temporal patterns.\\u003c/p\\u003e\\u003c/li\\u003e\\u003c/ul\\u003e\\u003c/p\\u003e\\u003cp\\u003eBoth models were implemented in TensorFlow/Keras, maintaining identical input-output structures for comparability.Hyperparameters tuned included hidden units, dropout rates, learning rates, batch sizes, and epochs. Four optimization methods were employed:\\u003c/p\\u003e\\u003cp\\u003e\\u003cul\\u003e\\u003cli\\u003e\\u003cp\\u003eGrid Search: Exhaustive but computationally intensive.\\u003c/p\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cp\\u003eBayesian Optimization: Probabilistic refinement with moderate efficiency.\\u003c/p\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cp\\u003eOptuna GridSampler: Adaptive structured search.\\u003c/p\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cp\\u003eOptuna TPE Sampler: Probabilistic adaptive strategy.\\u003c/p\\u003e\\u003c/li\\u003e\\u003c/ul\\u003e\\u003c/p\\u003e\\u003cp\\u003eEvaluation metrics included RMSE, directional accuracy (proportion of correctly predicted price movements), Sharpe ratios (measuring risk-adjusted returns), and computational cost (runtime and resource use).\\u003c/p\\u003e\\u003cp\\u003e\\u003c/p\\u003e\"},{\"header\":\"5. Results\",\"content\":\"\\u003cp\\u003eThe choice of HPO method significantly influenced both predictive accuracy and financial outcomes. Baseline LSTM models optimized with grid search yielded RMSE of 0.0068, directional accuracy of 54.2%, and Sharpe ratios of 0.70. Bayesian optimization provided marginal gains, raising accuracy to 57\\u0026ndash;59% but Sharpe ratios remained below 1.0, indicating weak financial viability. Optuna-based methods marked a decisive improvement. The GridSampler achieved 60% accuracy for both LSTM and CNN models, producing Sharpe ratios above 1.0 with lower computational cost than traditional methods. The TPE Sampler delivered the strongest results, raising LSTM accuracy to 63%, CNN accuracy to 61%, and Sharpe ratios above 1.2, all at minimal resource cost.\\u003c/p\\u003e\\u003cp\\u003e\\u003cdiv class=\\\"gridtable\\\"\\u003e\\u003ctable float=\\\"Yes\\\" id=\\\"Tab1\\\" border=\\\"1\\\"\\u003e\\u003ccaption language=\\\"En\\\"\\u003e\\u003cdiv class=\\\"CaptionNumber\\\"\\u003eTable 1\\u003c/div\\u003e\\u003cdiv class=\\\"CaptionContent\\\"\\u003e\\u003cp\\u003eHPO Technique and its impact\\u003c/p\\u003e\\u003c/div\\u003e\\u003c/caption\\u003e\\u003ccolgroup cols=\\\"5\\\"\\u003e\\u003cdiv align=\\\"left\\\" class=\\\"colspec\\\" colname=\\\"c1\\\" colnum=\\\"1\\\"\\u003e\\u003c/div\\u003e\\u003cdiv align=\\\"left\\\" class=\\\"colspec\\\" colname=\\\"c2\\\" colnum=\\\"2\\\"\\u003e\\u003c/div\\u003e\\u003cdiv align=\\\"left\\\" class=\\\"colspec\\\" colname=\\\"c3\\\" colnum=\\\"3\\\"\\u003e\\u003c/div\\u003e\\u003cdiv align=\\\"left\\\" class=\\\"colspec\\\" colname=\\\"c4\\\" colnum=\\\"4\\\"\\u003e\\u003c/div\\u003e\\u003cdiv align=\\\"left\\\" class=\\\"colspec\\\" colname=\\\"c5\\\" colnum=\\\"5\\\"\\u003e\\u003c/div\\u003e\\u003cthead\\u003e\\u003ctr\\u003e\\u003cth align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003eMethod\\u003c/p\\u003e\\u003c/th\\u003e\\u003cth align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003eDA (LSTM)\\u003c/p\\u003e\\u003c/th\\u003e\\u003cth align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003eDA (CNN)\\u003c/p\\u003e\\u003c/th\\u003e\\u003cth align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u003cp\\u003eSharpe Ratio\\u003c/p\\u003e\\u003c/th\\u003e\\u003cth align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u003cp\\u003eComputational Cost\\u003c/p\\u003e\\u003c/th\\u003e\\u003c/tr\\u003e\\u003c/thead\\u003e\\u003ctbody\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003eGrid Search\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003e58%\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003e56%\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u003cp\\u003eLow\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u003cp\\u003eHigh\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003eBayesian Optimisation\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003e59%\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003e57%\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u003cp\\u003eLow-Moderate\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u003cp\\u003eModerate\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003eOptuna GridSampler\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003e60%\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003e60%\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u003cp\\u003eModerate\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u003cp\\u003eLow\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003eOptuna TPESampler\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003e63%\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003e61%\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u003cp\\u003eHigh\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u003cp\\u003eLow\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003c/tbody\\u003e\\u003c/colgroup\\u003e\\u003c/table\\u003e\\u003c/div\\u003e\\u003c/p\\u003e\\u003cp\\u003eThe comparative results confirm that adaptive optimization not only improves predictive precision but also enhances portfolio profitability. Computational efficiency further underscores their practical relevance, with Optuna methods converging faster than traditional approaches.\\u003c/p\\u003e\"},{\"header\":\"6. Discussion\",\"content\":\"\\u003cp\\u003eThe findings underscore three key insights. First, statistical accuracy alone does not guarantee financial utility. Both grid search and Bayesian optimization improved error metrics but failed to deliver Sharpe ratios above unity, confirming that financial evaluation must go beyond RMSE.\\u003c/p\\u003e\\u003cp\\u003eSecond, Optuna\\u0026rsquo;s adaptive strategies, particularly TPE, demonstrated clear superiority. By dynamically balancing exploration and exploitation, TPE avoided premature convergence and identified hyperparameter configurations that maximized both predictive robustness and profitability. These results are especially relevant in volatile emerging markets such as India, where ETF dynamics are shaped by policy, liquidity, and global capital flows.Third, computational efficiency is not merely a technical concern but a strategic necessity. Models that recalibrate quickly enable near real-time deployment, reducing risk exposure and opportunity costs. Optuna\\u0026rsquo;s ability to combine higher accuracy with faster convergence makes it well-suited for both institutional and retail adoption. The broader implication is that HPO is not a peripheral task but a critical determinant of whether deep learning models succeed in financial applications. Adaptive probabilistic approaches elevate models from experimental tools to viable forecasting engines, bridging the gap between computational innovation and financial decision-making.\\u003c/p\\u003e\"},{\"header\":\"7. Conclusion\",\"content\":\"\\u003cp\\u003eThis study evaluated the impact of hyperparameter optimization on deep learning-based ETF prediction, focusing on the Nifty BeEs. By comparing grid search, Bayesian optimization, Optuna GridSampler, and Optuna TPE Sampler, the research demonstrated that optimization methods fundamentally alter predictive accuracy, financial relevance, and efficiency. Traditional methods yielded modest improvements but failed to align predictive performance with investment profitability. In contrast, Optuna-based approaches consistently outperformed. GridSampler delivered balanced gains in accuracy and Sharpe ratios above 1.0, while TPE achieved the highest accuracy (63% for LSTM, 61% for CNN) and Sharpe ratios above 1.2 at minimal computational cost.The findings highlight that hyperparameter optimization is not ancillary but central to deploying deep learning in finance. Adaptive methods enable models to transcend statistical benchmarks and deliver actionable financial value. For Indian ETFs, these results confirm that advanced optimization can transform predictive analytics into a practical tool for risk-adjusted returns.\\u003c/p\\u003e\\u003cp\\u003eThis research contributes to bridging computer science and financial econometrics, offering theoretical and practical insights. Future work could extend this framework to high-frequency data, cross-asset ETFs, or hybrid models incorporating macroeconomic indicators. Ultimately, integrating advanced HPO methods into financial modeling represents a decisive step toward unlocking the full potential of AI in capital markets.\\u003c/p\\u003e\"},{\"header\":\"Declarations\",\"content\":\"\\u003cp\\u003e\\u003cstrong\\u003eConflict of interest\\u003c/strong\\u003e\\u003cp\\u003eThe authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.\\u003c/p\\u003e\\u003c/p\\u003e\\u003ch2\\u003eFunding\\u003c/h2\\u003e\\u003cp\\u003eNo funding is received pertaining to this research.\\u003c/p\\u003e\\u003ch2\\u003eAuthor Contribution\\u003c/h2\\u003e\\u003cp\\u003eAlan Vellaiparambill wrote the main manuscript text and E.F. prepared figuresDr. Natchimuthu N reviewed the manuscript and supported the HPO process at ideation and implementation\\u003c/p\\u003e\\u003ch2\\u003eAcknowledgements\\u003c/h2\\u003e\\u003cp\\u003eThis work was supported by Department of Commerce, Christ Deemed to be University, India.\\u003c/p\\u003e\\u003ch2\\u003eData Availability\\u003c/h2\\u003e\\u003cp\\u003eThe data that support the findings of this study are openly available at the websites linked in the article.\\u003c/p\\u003e\"},{\"header\":\"References\",\"content\":\"\\u003col\\u003e\\u003cli\\u003e\\u003cspan\\u003eLevchenko O, Lemeshko O, Skorokhod Y, Khudov H, Khudov H (2024) Chain-structured neural architecture search for financial time series forecasting. J Comput Finance Analytics 18(2):101\\u0026ndash;120\\u003c/span\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cspan\\u003eKehinde O, Li Y, Zhang H (2025) Helformer: Attention-based deep learning for cryptocurrency price prediction. Expert Syst Appl 240:121245\\u003c/span\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cspan\\u003eGonzalvez J, Patel R, Singh A (2019) Bayesian optimization and Gaussian processes in financial modeling. Quant Finance 19(7):1125\\u0026ndash;1141\\u003c/span\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cspan\\u003eAkiba T, Sano S, Yanase T, Ohta T, Koyama M (2019) Optuna: A next-generation hyperparameter optimization framework. Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, 2623\\u0026ndash;2631\\u003c/span\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cspan\\u003eBergstra J, Bardenet R, Bengio Y, K\\u0026eacute;gl B (2011) Algorithms for hyper-parameter optimization. Adv Neural Inf Process Syst 24:2546\\u0026ndash;2554\\u003c/span\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cspan\\u003eBergstra J, Yamins D, Cox D (2013) Making a science of model search: Hyperparameter optimization in hundreds of dimensions for vision architectures. Proceedings of the 30th International Conference on Machine Learning, 115\\u0026ndash;123\\u003c/span\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cspan\\u003eZhang L, Wu J, Zhou D (2020) Beyond accuracy: Financial utility metrics in evaluating machine learning models. J Financial Data Sci 2(3):55\\u0026ndash;72\\u003c/span\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cspan\\u003eLi X, Hoi SC (2021) Evaluating financial machine learning models with utility-driven metrics. ACM Trans Manage Inform Syst 12(4):1\\u0026ndash;22\\u003c/span\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cspan\\u003eRamesh A, Nair S (2022) Deep learning adoption in Indian equity markets: Evidence from NIFTY 50. Asian J Finance Acc 14(1):42\\u0026ndash;61\\u003c/span\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cspan\\u003eGupta V, Iyer P (2023) Machine learning for derivatives pricing and ETF forecasting in India. Int J Financial Stud 11(2):33\\u0026ndash;49\\u003c/span\\u003e\\u003c/li\\u003e\\u003c/ol\\u003e\"}],\"fulltextSource\":\"\",\"fullText\":\"\",\"funders\":[],\"hasAdminPriorityOnWorkflow\":false,\"hasManuscriptDocX\":true,\"hasOptedInToPreprint\":true,\"hasPassedJournalQc\":\"\",\"hasAnyPriority\":true,\"hideJournal\":true,\"highlight\":\"\",\"institution\":\"\",\"isAcceptedByJournal\":false,\"isAuthorSuppliedPdf\":false,\"isDeskRejected\":\"\",\"isHiddenFromSearch\":false,\"isInQc\":false,\"isInWorkflow\":false,\"isPdf\":false,\"isPdfUpToDate\":true,\"isWithdrawnOrRetracted\":false,\"journal\":{\"display\":true,\"email\":\"info@researchsquare.com\",\"identity\":\"researchsquare\",\"isNatureJournal\":false,\"hasQc\":true,\"allowDirectSubmit\":true,\"externalIdentity\":\"\",\"sideBox\":\"\",\"snPcode\":\"\",\"submissionUrl\":\"/submission\",\"title\":\"Research Square\",\"twitterHandle\":\"researchsquare\",\"acdcEnabled\":true,\"dfaEnabled\":false,\"editorialSystem\":\"\",\"reportingPortfolio\":\"\",\"inReviewEnabled\":false,\"inReviewRevisionsEnabled\":true},\"keywords\":\"Deep learning, Index Prediction, Hyperparameter Optimization, Financial Forecasting\",\"lastPublishedDoi\":\"10.21203/rs.3.rs-7567095/v1\",\"lastPublishedDoiUrl\":\"https://doi.org/10.21203/rs.3.rs-7567095/v1\",\"license\":{\"name\":\"CC BY 4.0\",\"url\":\"https://creativecommons.org/licenses/by/4.0/\"},\"manuscriptAbstract\":\"\\u003cp\\u003eThe integration of deep learning into financial forecasting has significantly advanced predictive analytics, yet the effectiveness of these models is critically dependent on hyperparameter optimization (HPO). This study investigates the role of HPO in enhancing the predictive and financial performance of Long Short-Term Memory (LSTM) and one-dimensional Convolutional Neural Network (1D-CNN) models applied to the Nifty BeEs Exchange Traded Fund (ETF), a key proxy for the Indian equity market. Using daily log-return data from 2010 to 2025, four HPO techniques grid search, Bayesian optimization, Optuna GridSampler, and Optuna Tree-structured Parzen Estimator (TPE) were systematically compared. Evaluation metrics included Root Mean Squared Error (RMSE), directional accuracy (DA), Sharpe ratios, and computational cost. Results demonstrate that while traditional methods provide modest improvements, they fail to align statistical accuracy with financial viability. In contrast, Optuna-based approaches, particularly the TPE Sampler, significantly improved outcomes, raising LSTM accuracy to 63% and CNN accuracy to 61%, with Sharpe ratios exceeding 1.2 at minimal computational cost. These findings underscore that hyperparameter optimization is not a peripheral technical task but a strategic determinant of investment applicability, transforming deep learning models from theoretical constructs into practical forecasting engines. The study contributes to bridging methodological innovations in computer science with financial econometrics, offering actionable insights for ETF prediction in emerging markets.\\u003c/p\\u003e\",\"manuscriptTitle\":\"Impact of Hyperparameter Optimisation Techniques in Deep Learning-based Investment Predictions: An Indian ETF-based analysis\",\"msid\":\"\",\"msnumber\":\"\",\"nonDraftVersions\":[{\"code\":1,\"date\":\"2025-11-13 08:30:49\",\"doi\":\"10.21203/rs.3.rs-7567095/v1\",\"editorialEvents\":[{\"type\":\"communityComments\",\"content\":0}],\"status\":\"published\",\"journal\":{\"display\":true,\"email\":\"info@researchsquare.com\",\"identity\":\"researchsquare\",\"isNatureJournal\":false,\"hasQc\":true,\"allowDirectSubmit\":true,\"externalIdentity\":\"\",\"sideBox\":\"\",\"snPcode\":\"\",\"submissionUrl\":\"/submission\",\"title\":\"Research Square\",\"twitterHandle\":\"researchsquare\",\"acdcEnabled\":true,\"dfaEnabled\":false,\"editorialSystem\":\"\",\"reportingPortfolio\":\"\",\"inReviewEnabled\":false,\"inReviewRevisionsEnabled\":true}}],\"origin\":\"\",\"ownerIdentity\":\"ceb6d030-84fe-4859-9930-88f1f67ff1ba\",\"owner\":[],\"postedDate\":\"November 13th, 2025\",\"published\":true,\"recentEditorialEvents\":[],\"rejectedJournal\":[],\"revision\":\"\",\"amendment\":\"\",\"status\":\"posted\",\"subjectAreas\":[],\"tags\":[],\"updatedAt\":\"2025-11-23T16:38:43+00:00\",\"versionOfRecord\":[],\"versionCreatedAt\":\"2025-11-13 08:30:49\",\"video\":\"\",\"vorDoi\":\"\",\"vorDoiUrl\":\"\",\"workflowStages\":[]},\"version\":\"v1\",\"identity\":\"rs-7567095\",\"journalConfig\":\"researchsquare\"},\"__N_SSP\":true},\"page\":\"/article/[identity]/[[...version]]\",\"query\":{\"redirect\":\"/article/rs-7567095\",\"identity\":\"rs-7567095\",\"version\":[\"v1\"]},\"buildId\":\"XKTyCvWXoU3ODBz1xrDgd\",\"isFallback\":false,\"isExperimentalCompile\":false,\"dynamicIds\":[84888],\"gssp\":true,\"scriptLoader\":[]}","source_license":"CC-BY-4.0","license_restricted":false}