Strong denoising of financial time-series
Matthias J. Feiler
TL;DR
This paper presents a novel auto-encoder based method that uses mutual regularization through a conversational training process to significantly improve noise reduction in financial time-series data, revealing new regularities for trading.
Contribution
Introduces a mutual regularization approach with conversational training of auto-encoders to enhance denoising and discover new financial data regularities.
Findings
Enhanced signal-to-noise ratio in financial data
Discovery of new regularities in time-series
Potential for profitable trading strategies
Abstract
In this paper we introduce a method for significantly improving the signal to noise ratio in financial data. The approach relies on combining a target variable with different context variables and use auto-encoders (AEs) to learn reconstructions of the combined inputs. The objective is to obtain agreement among pairs of AEs which are trained on related but different inputs and for which they are forced to find common ground. The training process is set up as a "conversation" where the models take turns at producing a prediction (speaking) and reconciling own predictions with the output of the other AE (listening), until an agreement is reached. This leads to a new way of constraining the complexity of the data representation generated by the AE. Unlike standard regularization whose strength needs to be decided by the designer, the proposed mutual regularization uses the partner network…
Peer Reviews
Decision·Submitted to ICLR 2025
**Novel Approach to Denoising** The paper introduces a novel concept of using a "conversational" framework with two distinct AEs and translation layers for mutual regularization. This approach differs from traditional methods and offers a potentially powerful way to filter out noise by seeking agreement between independently trained AEs **Intriguing Application to Trading Strategy Discovery** The application of this denoising technique to discover trading strategies is a compelling demons
**Limited evaluation** The empirical results are based on a specific set of context variables and a limited time period. It's unclear how well this method generalizes to other markets, asset classes, or timeframes. More extensive experiments and robustness checks are needed to assess the generalizability of the findings. * The paper relies heavily on the profitability of the discovered trading strategies as evidence of successful denoising. While this is an interesting application, a more dir
The paper introduces an innovative regularization mechanism, where two AEs mutually regularize each other’s representations. This approach is novel and avoids the drawbacks of traditional regularization methods. The methodology is detailed, with a clear description of the interaction between the AEs and their translation layers. Additionally, empirical results on financial time-series demonstrate how the denoised signals can be applied to real-world trading scenarios. The paper is generally we
The dual-AE setup may introduce additional **computational overhead**, especially with high-dimensional data. The paper does not discuss scalability or computational trade-offs. The paper lacks **quantitative metrics** to objectively evaluate the performance of the denoising method. Only qualitative results are presented. Additionally, there is no **comparison with other state-of-the-art** denoising techniques, such as those based on wavelet-thresholding or stacked autoencoders. Including these
This work presents a new method to tackle the problem of time series prediction.
- I find some of the components of this paper not well-defined. Terms "listen" and "speak" are fine to use as long as they are well defined. I understand the intuition but cant verify if what I understand is correct. - Some components of this paper are not well-defined. Terms like "listen" and "speak" are acceptable as long as they are clearly defined. While I understand the intuition, I cannot verify if my understanding is correct. - There is a lack of quantitative results (I am not familiar
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
Taxonomy
TopicsComplex Systems and Time Series Analysis
MethodsSparse Evolutionary Training · Autoencoders