دانلود مقاله پیش‌بینی عملکرد شبکه‌های عصبی موجک و توپولوژی شبکه عصبی
چکیده
مقدمه
مروری بر مطالعات پیشین
فرضیه های تحقیق
روش‌شناسی، طراحی آزمایشی و توصیف داده‌ها
نتایج تجربی
بحث و مفاهیم
منابع
Abstract
Introduction
Literature review
Research hypotheses
Methodology, experimental design, and data description
Empirical results
Discussion and implications
References
چکیده
در این مطالعه، ما اثرات سودمند شبکه‌های عصبی را در ترکیب با توابع موجک بر عملکرد پیش‌بینی‌های بازار مالی تحلیل می‌کنیم. ما رویکردهای مختلف را در آزمایش‌های متعدد پیاده‌سازی می‌کنیم و توانایی‌های پیش‌بینی آن‌ها را با سری‌های زمانی مالی مختلف آزمایش می‌کنیم. ما به طور تجربی نشان می‌دهیم که هم شبکه‌های عصبی موجک و هم شبکه‌های عصبی با داده‌های از پیش پردازش‌شده توسط موجک‌ها از توپولوژی‌های شبکه کلاسیک بهتر عمل می‌کنند. با این حال، دقت پیش‌بینی‌های انجام‌شده در پیاده‌سازی الگوریتم‌های شبکه عصبی هنوز پتانسیل را برای اصلاح و بهبود بیشتر پیشنهاد می‌کند. از این رو، یافته‌های خود، مقایسه‌ها با استراتژی‌های «خرید و نگه‌داری» و ملاحظات اخلاقی را به طور انتقادی مورد بحث قرار می‌دهیم و چشم‌اندازهای آینده را به تفصیل بیان می‌کنیم.
توجه! این متن ترجمه ماشینی بوده و توسط مترجمین ای ترجمه، ترجمه نشده است.
Abstract
In this study, we analyse the advantageous effects of neural networks in combination with wavelet functions on the performance of financial market predictions. We implement different approaches in multiple experiments and test their predictive abilities with different financial time series. We demonstrate experimentally that both wavelet neural networks and neural networks with data pre-processed by wavelets outperform classical network topologies. However, the precision of conducted forecasts implementing neural network algorithms still propose potential for further refinement and enhancement. Hence, we discuss our findings, comparisons with “buy-and-hold” strategies and ethical considerations critically and elaborate on future prospects.
Introduction
Forecasting models tailored for financial time series are discussed frequently in business and science (Sezer et al., 2020). The development of computer-based methods has witnessed significant progress, which is well illustrated by, for example, Renaissance Technologies, led by James Simons. Renaissance Technologies has systematically outperformed market growth over many years through the execution and advanced analysis of algorithms and signals (Burton, 2016). However, the majority of actively managed funds1 based on statistical analysis display severe underperformance owing to lower yields earned compared to the respective benchmark (market) indices (Otuteye & Siddiquee, 2019). In particular, said underperformance renders itself visible once trading and management fees are considered, which are compared with passive investments, such as buy-and-hold strategies (Otuteye & Siddiquee, 2019).
Empirical results
Before elaborating on respective results, we note the stochastic nature of NNs to impose a great influence on the magnitude deviations between the predictions (Kaastra & Boyd, 1996). The before mentioned implication originates from the existence of many predictions, which enable the prediction of only small fractions of a percent from the intended value. In addition, however, a similar number of predictions deviate from the true value by several percent, according to nonoptimal gradient descent algorithms, as stated in Kaastra and Boyd (1996). In total, we record up to 600 predictions7 from each stock 7 20 times 30 days. or index and LSTM topology, less for shorter data sets,8 respectively. Furthermore, up to 20 predictions are added with NNs on MLP basis for each data set, thus, we obtain a total number of 58,446 predictions, which are documented and evaluated according to the back-test described previously. As suspected, the examination of the prediction results provides a differentiated picture of the quality of predictions regarding the different selected NNs. Therefore, it is favourable to address different perspectives within the respective evaluations. The first perspective envisages the results with a focus on the general performance of an NN over the time horizon from one to 30 days. Therefore, we consider all available results of each network, separated per time horizon but averaged across all data sets. The second perspective elucidates the performance within the predicted stocks and indices. For each stock or index, the best network is determined for each time horizon. Further, it can be deduced whether an NN performs better than others regarding certain data sets and time-horizons.