Paper: arXiv 2311.12129
Authors: Martin Winistörfer, Ivan Zhdankin
Abstract
Assessing the predictive power of both data and models holds paramount significance in time-series machine learning applications. Yet, preparing time series data accurately and employing an appropriate measure for predictive power seems to be a non-trivial task. This work involves reviewing and establishing the groundwork for a comprehensive analysis of shaping time-series data and evaluating various measures of dependence. Lastly, we present a method, framework, and a concrete example for selecting and evaluating a suitable measure of dependence.
Complexity vs Empirical Score
- Math Complexity: 6.5/10
- Empirical Rigor: 3.0/10
- Quadrant: Lab Rats — high math complexity, limited empirical testing
Why this score: The paper focuses on theoretical frameworks and mathematical definitions for measuring dependence in time series, with advanced discussions on tensor operations and statistical concepts, but lacks code, specific backtests, or datasets in the provided excerpt, placing it in the theoretical Lab Rats quadrant.
Research Flowchart
flowchart TD
A["Research Goal: <br>Find Optimal Dependence Measure for Financial Time-Series"] --> B["Data Input: <br>Cross-Asset Financial Data"]
B --> C["Methodology: <br>Review & Establish Data Shaping Framework"]
C --> D["Computational Process: <br>Evaluate Measures of Dependence"]
D --> E{"Measure Selection & <br>Evaluation Loop"}
E --> D
E --> F["Outcomes: <br>1. Concrete Framework for Selection<br>2. Comparative Analysis of Measures<br>3. Final Recommendation"]
Evidence Card
What the authors claim, on what data, and who has checked it. Read from the paper's own text by a language model: every quote is verbatim, nothing was re-run. How to read this card
Proposes
The paper proposes a method and framework, with a concrete example, for selecting and evaluating a suitable measure of dependence for financial time‑series.
Source sentence
Lastly, we present a method, framework, and a concrete example for selecting and evaluating a suitable measure of dependence.
Assumes 5
- Predictive power must be assessed over a time horizon that captures regimes persisting into the future. data
Source sentence
Therefore, in this paper, we always consider a time horizon with the ultimate goal of detecting a regime that persists over time from historical data through future data.
- The measure of dependence will use a simple statistical model that differs from the production model. model
Source sentence
Another crucial finding is that, irrespective of the chosen measure of dependence, the model based on the measure of dependence will always differ from the production model.
- Sample size for window calculations should not be smaller than 100 observations. statistical
Source sentence
Based on experience, the sample size should not be smaller than 100.
- Iterative verification using production model scores can identify effective measures and features. model
Source sentence
It is assumed that this is an iterative process, and after a certain number of iterations, good measures and features can be distinguished from the less effective ones.
Show 1 more
- Synthetic data can be used as a control dataset where predictive power is known in advance. data
Source sentence
Purely random data can be used as a control dataset. In this setup, the data with high and low predictive power is already known in advance.
Data
No dataset named in the text.
Costs
Not applicable
Finds 3
- Careful preparation of time‑series data aligned with the chosen measure of dependence is essential for uncovering predictive power. theoretical
Source sentence
It is essential that time-series data align with the chosen measure of dependence and the specific problem statement.
- Correlating historical data with future data without information leakage enables a measure of dependence to genuinely uncover predictive power. predictive-accuracy
Source sentence
Conversely, correlating historical data with future data, without information leakage, enables a measure of dependence to genuinely uncover predictive power.
- Regardless of the measure used, the dependence model will differ from the production model, typically being simpler than deep‑learning based production models. theoretical
Source sentence
Typically, a measure of dependence employs a simple module, like a statistical model. In contrast, a production model utilizes a more complex model, such as a deep learning network.
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 12 related papers). Coverage is partial, so this is neither replication nor refutation.
Code
No repository by the authors found in the paper.
Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.