Nowcasting Imports from Customs Transaction Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current predictive analytics face challenges in accurately determining import-related data due to time delays, incompleteness, and errors in data reporting, which hinders understanding of actual imports, import categories, and overall imports from countries of interest.
Innovation Solution
A method that defines a signal relationship between granular data and target data, aggregates granular data in a hierarchical format, processes it to correct errors and fill gaps, and applies time series transformations and machine learning models to estimate target data values, providing nowcasting predictions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If data reporting is performed using traditional census methods, then data completeness is improved, but time delay increases
Solution Approach 1:
The patent segments the data reporting process into two parallel streams: (1) traditional census data collection for completeness, and (2) granular transaction-level data collection for timeliness. By dividing the data sources and processing paths, the system achieves both complete information coverage and rapid time response without requiring one method to compensate for the other's weaknesses.
Solution Approach 2:
The patent introduces an intermediary reconciliation process that mediates between granular transaction data and aggregated census data. This intermediary layer validates, correlates, and harmonizes the two data streams, allowing the system to leverage the timeliness of transaction data while maintaining the completeness assurance of census data through cross-validation.
2Loss of time
If granular transaction data is used for nowcasting, then time response is improved, but data accuracy deteriorates due to errors and redactions
Solution Approach 1:
The patent implements a feedback mechanism where census data (which is more accurate but delayed) is used to validate and correct granular transaction data (which is timely but less accurate). The system continuously compares transaction-level data against census aggregates, identifies discrepancies, and applies corrections to the granular data, thereby maintaining measurement precision while preserving the time response advantage.
Solution Approach 2:
The patent performs preliminary validation and error correction on granular transaction data before it is used for nowcasting. By pre-processing the granular data to identify and correct common errors (such as redacted fields, formatting issues, and obvious anomalies) before aggregation, the system improves data accuracy in advance, reducing the need for extensive post-processing and enhancing the reliability of time-sensitive predictions.
3Productivity
If hierarchical aggregation is applied to granular data, then processing efficiency is improved, but data detail is lost
Solution Approach 1:
The patent implements a nested hierarchical aggregation structure where granular transaction data is aggregated at multiple levels (transaction → daily → weekly → monthly → yearly) while preserving the ability to drill down from any higher level to the underlying detailed data. Each aggregation level contains references to the next lower level, creating a nested structure that enables efficient processing at the required level of detail while maintaining access to full granularity when needed for analysis or validation.
Data Source
AI summary
A signal relationship is defined between a granular data value and a target data value. At least a portion of the granular data value corresponds to a granular latency value that is smaller than a target data latency value corresponding to the target data value. Granular data corresponding to the granular data value is interpreted. The granular data is aggregated in response to the signal relationship. A value of the target data value for a selected time reference is estimated, and the estimated value of the target data value is provided as a nowcasting prediction of the target data value.