pmm-research

name	pmm-research
description	Research assistant for writing academic papers about Polynomial Maximization Method (PMM) - parameter estimation with asymmetric non-Gaussian distributions

PMM Research & Paper Writing Skill

Comprehensive assistance for conducting research and writing academic papers using the Polynomial Maximization Method (МПП - Метод Максимізації Поліномів) for parameter estimation in statistical models with asymmetric non-Gaussian error distributions.

When to Use This Skill

This skill should be triggered when:

• Writing academic papers about PMM methodology
• Implementing PMM estimators for regression, time series, or other statistical models
• Conducting Monte Carlo simulations to evaluate PMM performance
• Comparing PMM with classical methods (OLS, MLE, CSS)
• Preparing reproducibility packages for journal submissions
• Creating statistical analysis with asymmetric error distributions
• Developing R packages for PMM-based estimation
• Working on papers about variance reduction in parameter estimation

Key Concepts

Polynomial Maximization Method (PMM)

• Core Principle: Uses polynomials of degree S to estimate distribution parameters
• PMM1 (S=1): Equivalent to classical Ordinary Least Squares (OLS)
• PMM2 (S=2): Provides variance reduction for asymmetric distributions
• Variance Reduction Formula: g = 1 - c₃² / (2 + c₄) where c₃ is skewness, c₄ is kurtosis
• Original Developer: Yu. P. Kunchenko (1992)

Application Domains

1. Linear Regression - Parameter estimation with non-Gaussian errors
2. Time Series (ARIMA) - AR, MA, ARMA, ARIMA models with asymmetric innovations
3. Autoregressive Models - AR(p) with non-normal residuals
4. Moving Average Models - MA(q) with skewed errors
5. Economic/Financial Data - WTI crude oil prices, stock returns, volatility modeling

Performance Characteristics

Distribution	Skewness	Kurtosis	Theoretical Efficiency	Actual Improvement
Gamma (α=0.5)	2.83	12	57%	~50%
Exponential	2.00	6	50%	~45%
Gamma (α=2)	1.41	3	40%	~35%
Lognormal	1.00	1.5	29%	~25%

Quick Reference

PMM2 Linear Regression (R)

library(EstemPMM)

# 1. Create data with asymmetric errors
n <- 100
x <- rnorm(n)
errors <- rgamma(n, shape = 2, scale = 1) - 2
y <- 2 + 1.5 * x + errors
data <- data.frame(x = x, y = y)

# 2. Fit model using PMM2
fit <- lm_pmm2(y ~ x, data = data)

# 3. Review results and compare with OLS
summary(fit)
compare_with_ols(y ~ x, data)

# 4. Check variance reduction
vf <- pmm2_variance_factor(fit@m2, fit@m3, fit@m4)
cat("Variance reduction factor g:", vf$g, "\n")

PMM2 ARIMA Models (R)

library(EstemPMM)

# AR(1) model with PMM2
ar_fit <- ar_pmm2(y, order = 1, method = "pmm2")

# ARMA(1,1) model
arma_fit <- arma_pmm2(y, order = c(1, 1))

# ARIMA(1,1,1) model
arima_fit <- arima_pmm2(y, order = c(1, 1, 1))

# Compare methods
compare_ar_methods(y, p = 1)

Monte Carlo Simulation Template

# Monte Carlo simulation for PMM2 performance evaluation
run_simulation <- function(n = 100, reps = 1000, dist = "gamma") {
  results <- matrix(NA, nrow = reps, ncol = 4)
  colnames(results) <- c("OLS_beta", "PMM2_beta", "OLS_var", "PMM2_var")

  for (i in 1:reps) {
    # Generate data
    x <- rnorm(n)
    if (dist == "gamma") {
      errors <- rgamma(n, shape = 2) - 2
    } else if (dist == "exp") {
      errors <- rexp(n, rate = 1) - 1
    }
    y <- 1 + 0.5 * x + errors

    # Fit models
    ols <- lm(y ~ x)
    pmm2 <- lm_pmm2(y ~ x, data = data.frame(x, y))

    # Store results
    results[i, 1] <- coef(ols)[2]
    results[i, 2] <- coef(pmm2)[2]
    results[i, 3] <- vcov(ols)[2, 2]
    results[i, 4] <- vcov(pmm2)[2, 2]
  }

  return(results)
}

# Run and analyze
results <- run_simulation(n = 200, reps = 2000, dist = "gamma")
variance_ratio <- mean(results[, 4]) / mean(results[, 3])
cat("Empirical variance ratio (PMM2/OLS):", variance_ratio, "\n")

Paper Structure Template

# Paper: Applying PMM to [Your Model Type]

## 1. Introduction
- Motivation: Why asymmetric errors matter
- Research gap: What existing methods miss
- Contribution: How PMM improves estimation

## 2. Methodology
### 2.1 Polynomial Maximization Method
- Theoretical foundation (Kunchenko, 1992)
- Variance reduction formula
- Adaptive estimation procedure

### 2.2 Model Specification
- [Your specific model: ARIMA, regression, etc.]
- Assumptions and conditions
- Parameter space

### 2.3 Implementation
- Algorithm description
- Computational complexity
- R/Python implementation

## 3. Monte Carlo Study
### 3.1 Simulation Design
- Sample sizes: n = {50, 100, 200, 500}
- Replications: 1000-5000
- Distributions: Gamma, Exponential, Lognormal

### 3.2 Performance Metrics
- Bias: E(θ̂) - θ
- Variance: Var(θ̂)
- MSE: E[(θ̂ - θ)²]
- Relative efficiency: Var(OLS)/Var(PMM2)

## 4. Empirical Application
- Real dataset description
- Diagnostic tests (stationarity, residual analysis)
- Model comparison (AIC, BIC, RMSE)
- Out-of-sample forecasting

## 5. Results
- Tables: Parameter estimates, standard errors, efficiency gains
- Figures: Convergence plots, residual diagnostics, QQ-plots

## 6. Conclusion
- Summary of findings
- Practical implications
- Future research directions

Reference Files

Available Documentation

• references/README.md - PMM2-ARIMA reproducibility package guide
• references/releases.md - Version history and release notes (2 releases)
• references/file_structure.md - Complete repository structure (90 items)

Key Resources from References

1. EstemPMM R Package - Production-ready implementation
2. PMM2-ARIMA Repository - ARIMA models with PMM2 estimator
3. Monte Carlo Framework - 128,000+ simulation template
4. WTI Case Study - Real-world application example
5. LaTeX Templates - Article formatting and structure

Working with This Skill

For New Papers

Step 1: Define Research Question

- What model are you estimating? (regression, ARIMA, GLM, etc.)
- What type of asymmetry do you expect? (right-skewed, left-skewed)
- What is your comparison baseline? (OLS, MLE, CSS-ML)

Step 2: Implement PMM Estimator

# Start with EstemPMM package for regression
library(EstemPMM)

# Or adapt PMM2-ARIMA code for time series
source("scripts/pmm2_estimation.R")

# Or implement from scratch using cumulant formulas
# See references/README.md for mathematical details

Step 3: Design Monte Carlo Study

# Use template above
# Vary: sample size, distribution, parameters
# Replications: 1000+ for publication quality
# Seed: Set for reproducibility (e.g., seed = 12345)

Step 4: Find Real Dataset

# Economic: GDP, inflation, oil prices (FRED database)
# Financial: Stock returns, volatility (Yahoo Finance)
# Environmental: Temperature, rainfall (NOAA)
# Medical: Survival times, biomarkers

Step 5: Create Reproducibility Package

data/           # Raw datasets
scripts/        # R scripts for all analyses
results/        # Auto-generated outputs
EstemPMM-lib/   # Bundled package version
README.md       # Replication instructions
DESCRIPTION     # R dependencies

For Code Development

R Package Structure

R/
  ├── pmm2_main.R         # Core estimation functions
  ├── pmm2_classes.R      # S4 class definitions
  ├── pmm2_utils.R        # Helper functions
  ├── pmm2_inference.R    # Bootstrap/asymptotic inference
  ├── pmm2_diagnostics.R  # Model diagnostics
  └── pmm2_simulation.R   # Monte Carlo utilities

man/                      # Roxygen2 documentation
tests/                    # Unit tests
vignettes/                # Tutorial vignettes

Essential Functions to Implement

1. pmm2_fit() - Core estimator
2. pmm2_vcov() - Variance-covariance matrix
3. pmm2_cumulants() - Sample cumulant estimation
4. pmm2_variance_factor() - Theoretical efficiency g
5. compare_methods() - PMM vs OLS/MLE comparison

For Paper Writing

Required Sections

1. Abstract (150-250 words)

   • Problem, method, results, conclusion

2. Introduction (3-4 pages)

   • Motivation with real examples
   • Literature review (cite Kunchenko, Zabolotnii papers)
   • Research gap and contribution

3. Methodology (4-6 pages)

   • PMM theoretical foundation
   • Your specific model adaptation
   • Computational algorithm

4. Simulations (3-5 pages)

   • Design justification
   • Results tables (bias, variance, MSE)
   • Convergence and robustness checks

5. Application (3-4 pages)

   • Dataset description and diagnostics
   • Estimation results
   • Model comparison
   • Forecasting performance

6. Conclusion (1-2 pages)

   • Summary and practical implications
   • Limitations and extensions

Key Citations

@article{kunchenko1992,
  author = {Kunchenko, Yu. P. and Lega, Yu. G.},
  title = {Polynomial parameter estimation of close to Gaussian random variables},
  journal = {Radioelectronics and Communications Systems},
  year = {1992}
}

@inproceedings{zabolotnii2018pmm,
  author = {Zabolotnii, S. and Warsza, Z.L. and Tkachenko, O.},
  title = {Polynomial Estimation of Linear Regression Parameters for the Asymmetric PDF of Errors},
  booktitle = {Automation 2018},
  publisher = {Springer},
  year = {2018},
  doi = {10.1007/978-3-319-77179-3_75}
}

@article{zabolotnii2025arima,
  author = {Zabolotnii, Serhii},
  title = {Applying the Polynomial Maximization Method to Estimate ARIMA Models with Asymmetric Non-Gaussian Innovations},
  journal = {Submitted for publication},
  year = {2025},
  doi = {10.5281/zenodo.17529589}
}

Common Patterns

Pattern 1: Quick Efficiency Check

# Before full Monte Carlo, check theoretical efficiency
c3 <- 1.5  # Your estimated skewness
c4 <- 3.0  # Your estimated kurtosis
g <- 1 - c3^2 / (2 + c4)
cat("Theoretical variance reduction:", (1 - g) * 100, "%\n")
# If g < 0.8, PMM2 should provide >20% variance reduction

Pattern 2: Reproducibility Workflow

# Master script: scripts/run_full_study.R
source("scripts/00_setup.R")              # Load packages
source("scripts/01_data_preparation.R")   # Import/clean data
source("scripts/02_estimation.R")         # Fit all models
source("scripts/03_monte_carlo.R")        # Run simulations
source("scripts/04_create_tables.R")      # Generate LaTeX tables
source("scripts/05_create_figures.R")     # Create publication plots
source("scripts/06_generate_report.R")    # Compile narrative report

Pattern 3: Model Comparison Table

# Create comparison table for paper
results_df <- data.frame(
  Model = c("OLS", "PMM2", "MLE"),
  Estimate = c(ols_est, pmm2_est, mle_est),
  SE = c(ols_se, pmm2_se, mle_se),
  AIC = c(ols_aic, pmm2_aic, mle_aic),
  BIC = c(ols_bic, pmm2_bic, mle_bic)
)

# Export to LaTeX
library(xtable)
print(xtable(results_df, digits = 4),
      file = "results/method_comparison.tex")

Best Practices

Research Workflow

1. ✅ Set seed for all random operations (reproducibility)
2. ✅ Version control with Git (track all changes)
3. ✅ Document dependencies in DESCRIPTION/requirements.txt
4. ✅ Bundle package versions (e.g., EstemPMM v0.1.1)
5. ✅ Archive on Zenodo for DOI and long-term preservation
6. ✅ Share code publicly (GitHub) before submission

Statistical Rigor

1. ✅ Test stationarity for time series (ADF, KPSS tests)
2. ✅ Check residuals (normality, autocorrelation, heteroscedasticity)
3. ✅ Report all methods (CSS, CSS-ML, PMM2 for ARIMA)
4. ✅ Use multiple metrics (AIC, BIC, RMSE, MAE)
5. ✅ Out-of-sample validation (not just in-sample fit)
6. ✅ Bootstrap inference when analytical se not available

Code Quality

1. ✅ Modular functions (one function = one task)
2. ✅ Roxygen documentation for all exported functions
3. ✅ Unit tests with testthat (>80% coverage)
4. ✅ Clear variable names (no x1, temp, res2)
5. ✅ Vectorized operations (avoid explicit loops when possible)
6. ✅ Progress bars for long-running simulations

Resources

Official Repositories

• EstemPMM: https://github.com/SZabolotnii/EstemPMM (CRAN package)
• PMM2-ARIMA: https://github.com/SZabolotnii/PMM2-ARIMA (Reproducibility package)

Key Papers

1. Kunchenko & Lega (1992) - Original PMM formulation
2. Zabolotnii et al. (2018) - PMM for asymmetric regression errors
3. Zabolotnii et al. (2022) - PMM for autoregressive models
4. Zabolotnii et al. (2023) - PMM for moving average models
5. Zabolotnii (2025) - PMM for ARIMA models (under review)

Datasets for Examples

• WTI Crude Oil Prices: FRED database (DCOILWTICO)
• Auto MPG: UCI Machine Learning Repository
• Financial Returns: Yahoo Finance API
• Economic Indicators: FRED, World Bank, Eurostat

Notes

• PMM2 is most effective when skewness |c₃| > 1.0
• For symmetric distributions (c₃ ≈ 0), PMM2 ≈ OLS
• Computational cost is similar to OLS (no major overhead)
• Bootstrap recommended for small samples (n < 50)
• Always report both OLS and PMM2 for transparency

Updating This Skill

To enhance with new research:

1. Add new PMM applications (GLM, survival analysis, etc.)
2. Update with published papers and citations
3. Include new simulation scenarios
4. Add real-world case studies
5. Expand code templates for other languages (Python, Julia)

---

Generated by Skill Seeker | Optimized for PMM Research & Academic Writing