[Paper Review] A Time-Scale Variational Approach to Inflation, Unemployment and Social Loss

Monika Dryl, Agnieszka B. Malinowska+1 more|arXiv (Cornell University)|Apr 18, 2013

Labor market dynamics and wage inequality11 references3 citations

TL;DR

This paper introduces a time-scale calculus framework to model inflation, unemployment, and social loss, unifying continuous and discrete approaches. By optimizing over time scales (specifically $ h\mathbb{Z} $), it achieves significantly better empirical fit than classical models, showing that weekly data sampling ($ h \approx 0.11 $–$ 0.22 $) minimizes social loss functional error, thus validating the need for higher-frequency data collection in macroeconomic modeling.

ABSTRACT

Both inflation and unemployment inflict social losses. When a tradeoff exists between the two, what would be the best combination of inflation and unemployment? A well known approach in economics to address this question consists to write the social loss as a function of the rate of inflation $p$ and the rate of unemployment $u$, with different weights, and then, using known relations between $p$, $u$, and the expected rate of inflation $π$, to rewrite the social loss function as a function of $π$. The answer is achieved by applying the theory of the calculus of variations in order to find an optimal path $π$ that minimizes the total social loss over a given time interval. Economists dealing with this question use a continuous or a discrete variational problem. Here we propose to use a time-scale model, unifying available results in the literature. Moreover, the new formalism allow us to obtain new insights to the classical models when applied to real data of inflation and unemployment.

Motivation & Objective

• To unify continuous and discrete models of inflation and unemployment into a single time-scale framework.
• To address the limitations of classical models in approximating real-world social loss data.
• To determine the optimal data sampling frequency that minimizes the discrepancy between theoretical and empirical social loss functionals.
• To demonstrate that classical models can be accurate if paired with appropriate time-scale sampling, rather than being inherently flawed.
• To provide a new variational framework on time scales that allows for dynamic optimization of expected inflation paths under social loss minimization.

Proposed method

• Formulates a dynamic social loss functional on time scales $ \mathbb{T} = h\mathbb{Z} $, generalizing continuous and discrete models.
• Applies the calculus of variations on time scales to derive necessary and sufficient optimality conditions for minimizing total social loss.
• Derives an explicit solution for the optimal expected inflation path $ \pi^* $ on $ h\mathbb{Z} $ using delta derivatives and variational principles.
• Calibrates the time scale parameter $ h $ by minimizing the relative error between the time-scale model's functional $ \Lambda_h $ and empirical data $ \Lambda_E $.
• Uses real U.S. data (2000–2010) for inflation and unemployment to compute and compare functionals across continuous, discrete, and time-scale models.
• Employs numerical optimization to identify the best $ h $ (e.g., $ h = 0.11 $, $ 0.22 $) that minimizes error in the time-scale model.

Experimental results

Research questions

• RQ1 Can a unified time-scale model improve the accuracy of social loss functionals compared to classical continuous and discrete models?
• RQ2 What is the optimal data sampling frequency (i.e., graininess $ h $) that minimizes the error between theoretical and empirical social loss?
• RQ3 Why do classical continuous and discrete models fail to approximate real data well, and can this be corrected by adjusting the sampling interval?
• RQ4 Does the time-scale approach allow for better prediction of optimal inflation paths under social loss minimization?
• RQ5 Can the classical models achieve comparable accuracy if paired with higher-frequency data, and if so, what is the required sampling rate?

Key findings

• The time-scale model with $ h = 0.11 $ or $ h = 0.22 $ (approximately weekly sampling) yields the smallest relative error between the theoretical and empirical social loss functionals.
• The relative error between $ \Lambda_h $ and $ \Lambda_E $ is reduced to 0.33% in 2009 and 0.58% in 2004, with an average 11-year error of 7.16% for the time-scale model.
• In contrast, the continuous model has an average 11-year relative error of 97.32%, and the discrete model exceeds 4×10^90, indicating poor empirical fit.
• The classical models fail not due to theoretical flaws, but due to inappropriate sampling frequency; their accuracy improves significantly when paired with optimal $ h $.
• The optimal $ h $ values (0.11–0.22) correspond to weekly data collection, suggesting that higher-frequency data collection is critical for accurate macroeconomic modeling.
• The time-scale model provides a mathematically rigorous and empirically validated framework that unifies and improves upon existing continuous and discrete approaches.