Paper: arXiv 2404.08136
Authors: Eric Luxenberg, Stephen Boyd
Abstract
An exponentially weighted moving model (EWMM) for a vector time series fits a new data model each time period, based on an exponentially fading loss function on past observed data. The well known and widely used exponentially weighted moving average (EWMA) is a special case that estimates the mean using a square loss function. For quadratic loss functions EWMMs can be fit using a simple recursion that updates the parameters of a quadratic function. For other loss functions, the entire past history must be stored, and the fitting problem grows in size as time increases. We propose a general method for computing an approximation of EWMM, which requires storing only a window of a fixed number of past samples, and uses an additional quadratic term to approximate the loss associated with the data before the window. This approximate EWMM relies on convex optimization, and solves problems that do not grow with time. We compare the estimates produced by our approximation with the estimates from the exact EWMM method.
Complexity vs Empirical Score
- Math Complexity: 7.5/10
- Empirical Rigor: 3.5/10
- Quadrant: Lab Rats — high math complexity, limited empirical testing
Why this score: The paper introduces a convex optimization framework for exponentially weighted moving models, involving advanced concepts like convex duality and quadratic approximations, indicating high mathematical density. While the examples discuss practical applications like covariance estimation and quantile regression, the paper lacks backtests, specific datasets, or implementation details for quantitative finance, resulting in lower empirical rigor.
Research Flowchart
flowchart TD A["Research Goal: EWMM Approximation"] --> B["Methodology: Window + Quadratic Approx"] B --> C["Data: Vector Time Series"] C --> D["Process: Solve Convex Opt"] D --> E["Exact EWMM"] D --> F["Approximate EWMM"] E --> G["High Compute + Storage"] F --> H["Fixed Storage + Efficient Compute"] G & H --> I["Outcome: Validated Approximation"]
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 an approximate recursive method for EWMM that stores only a fixed-size window of past samples and solves a constant-size convex optimization problem.
Source sentence
We propose a general method for computing an approximation of EWMM, which requires storing only a window of a fixed number of past samples, and uses an additional quadratic term to approximate the loss associated with the data before the window. This approximate EWMM relies on convex optimization, and solves problems that do not grow with time.
Assumes 3
- The loss function ℓ(x;θ) is convex in θ and the parameter set Θ is convex. model
Source sentence
We consider a model of the data x ∈Rn that is parametrized by θ ∈Θ ⊆Rm, and specified by the loss function ℓ: Rn × Θ →R, which we assume is convex in θ. (In particular, we assume Θ is a convex set.)
- There exists at least one minimizer for the EWMM optimization problem. model
Source sentence
We will assume that there is at least one minimizer in the argmin above; if there are multiple minimizers, we can simply choose one.
- The user wishes to use an EWMM and the paper does not address whether EWMM is appropriate for a given application. other
Source sentence
We simply assume that a user wishes to use an EWMM, and give a practical method to carry out the evaluation with memory and computation that do not grow with time.
Data
- 10 Industry Portfolio dataset market-prices public
Source sentence
We use the 10 Industry Portfolio dataset from Kenneth French’s data library [Fre24]. The dataset contains daily returns for 10 value-weighted industry portfolios.
Costs
Not applicable
Finds 3
- The approximate finite‑memory EWMM produces estimates that are close to those of the exact EWMM. theoretical
Source sentence
We demonstrate our approximate recursive method with examples, showing that it finds models that are close to the exact EWMM models, fit using all past data. We see that the approximate finite memory EWMM is able to closely match the performance from the exact EWMM.
- The recursive formulation yields linear growth in computation time, whereas the naïve method grows quadratically. performance
Source sentence
As expected the na¨ıve method, which saves all past data and directly computes the estimate using all past value, grows quadratically in time, whereas the recursive method grows linearly.
- The sparse inverse covariance estimates vary considerably over time, reflecting changing market conditions. statistical
Source sentence
These plots show that the sparse inverse covariance estimate varies considerably with time, i.e., market conditions.
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 11 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.
Cited References (46)
- CVXPY : A P ython-embedded modeling language for convex optimization (2016)
- Convex Optimization (2004)
- Portfolio Selection (1952)
- Convex analysis (1970)
- Convex Analysis (2006)
- Large-Scale Convex Optimization: Algorithms & Analyses via Monotone Operators (2022)
- CVX : M atlab software for disciplined convex programming, version 2.1 (2014)
- CVXR : An R Package for Disciplined Convex Optimization (2020) — link
and 38 more
- Convex Optimization in J ulia (2014) — link
- Disciplined convex programming (2006)
- The P2 algorithm for dynamic calculation of quantiles and histograms without storing observations (1985)
- Exponential smoothing for predicting demand (1956) — link
- A survey of algorithms and analysis for adaptive online learning (2017)
- Short-Term Load Forecasting Using General Exponential Smoothing (1971) — link
- Regression quantiles (1978)
- Sparse inverse covariance estimation with the L asso (2007) — link
- Matrix computations (1989)
- Regression Shrinkage and Selection via the Lasso (1996)
- French Data Library (2024)
- Sequential quadratic programming (1995)
- Robust locally weighted regression and smoothing scatterplots (1979)
- Space-efficient online computation of quantile summaries (2001)
- Sigma-point kalman filters for probabilistic inference in dynamic state-space models (2004)
- On the distribution of points in a cube and the approximate evaluation of integrals (1967)
- Discrete-time Signal Processing (1999)
- Introduction to online convex optimization (2016)
- Algorithms for Optimization (2019)
- Time Series Analysis: Forecasting and Control (2015)
- Analysis of Financial Time Series (2002)
- Robust Estimation of a Location Parameter (1992) — link
- The E lements of S tatistical L earning: D ata M ining, I nference, and P rediction (2001)
- Least-squares estimation: from Gauss to Kalman (1970) — link
- Theory of the C ombination of O bservations L east S ubject to E rrors (1995)
- Fitting Laplacian regularized stratified Gaussian models (2021)
- On distributions admitting a sufficient statistic (1936)
- Sufficient statistics and intrinsic accuracy (1936)
- Sur les lois de probabilites estimation exhaustive (1935)
- Performance bounds for linear stochastic control (2009)
- Approximate dynamic programming via iterated Bellman inequalities (2015)
- Quadratic approximate dynamic programming for input-affine systems (2014)
- Algorithms for Decision Making (2022)
- Local likelihood estimation (1987)
- The Barra US equity model (USE4) , methodology notes (2011)
- Approximate Dynamic Programming: Solving the Curses of Dimensionality (2011)
- A Simple Method for Predicting Covariance Matrices of Financial Returns (2023)
- Abstract Dynamic Programming (2022)
0 of 46 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.