Multiple imputation methods: a case study of daily gold price.

Area of Science:

• Financial econometrics
• Data science
• Statistical modeling

Background:

• Missing values are a common challenge in financial time-series data collection.
• Accurate data imputation is crucial for reliable financial analysis and forecasting.

Purpose of the Study:

• To compare the effectiveness of various data imputation techniques for financial time-series data.
• To identify the most robust imputation method for daily gold price data.
• To assess the impact of missing data proportion on imputation accuracy.

Main Methods:

• Evaluation of mean imputation, k-nearest neighbor (KNN), hot deck, random forest, support vector machine (SVM), and spline imputation.
• Validation using actual daily gold closing prices.
• Performance assessment using metrics: mean error (ME), mean absolute error (MAE), root mean square error (RMSE), mean percentage error (MPE), and mean absolute percentage error (MAPE).

Main Results:

• K-nearest neighbor (KNN) imputation demonstrated superior performance across all evaluated accuracy metrics.
• The predictive accuracy of all tested imputation methods decreased as the percentage of missing data increased.
• Spline imputation and random forest showed moderate performance, while mean imputation and hot deck were less effective.

Conclusions:

• K-nearest neighbor (KNN) is recommended as a highly effective and dependable method for imputing missing values in financial time-series data.
• The proportion of missing data significantly impacts the precision of imputation techniques.
• Further research could explore hybrid imputation methods or advanced machine learning models for complex financial datasets.