Similarity searching modell with Excel Zoltn Varga Ph

Similarity searching modell with Excel Zoltán Varga Ph. D student SZIU

Previous (Classical) model for sales forecast 0. Historical sales data 6. Interpretation to the actual data 1. Periodical simple average 2. Smooth step 1 by moving average 3. Smooth step 2 by exponential cleaning 4. Estimation 5. Forecasted values

Similarity analysis (COCO – Component-based Object Comparison for Objectivity) 0. Raw OAM (Object-Attribute Matrix 6. Optimizing 5. Definition of model error for minimizing 1. Ranked OAM 2. Definition of stair cases (range of results) 3. Definition of difference of neighboring stairs (restrictions) 4. Definition of object function

The story that lead to the birth of similarity searching model • EMC 2014 • Katalin Óhegyi and stock market dataset • Philosophy of similarity by COCO

Similarity searching model 0. Times series 6. Evaluating to the actual data by category accuracy and directional stability 1. tn/tn+1; tn+1/tn+2… ratio 5. First elements of the sequences m+1, 2 , 3…z are stored until z 2. Generation of non-overlapping categories 3. Generation of sequences (5 category points at most) 4. Searching for the most similar pair (m) of the last known sequence (n), and repeated search for the sequence m+1 with its first element is stored

Test on Crude Futures Open • 184 weeks used to forecast the next 25 • 19 categories Category difference 25 cases 100, 00% 0 1 2 3 4 5 6 7 8 9 4 6 8 4 1 0 0 1 16% 24% 32% 16% 4% 0% 0% 4% Directional accuracy: 18 Directional stability: 72%

Test on Crude Futures Open - Benchmark • 184 weeks used to find the most common Quantity of cathegories 30 25 20 15 10 5 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19

Test on Crude Futures Open – Benchmark results • 184 weeks used to forecast the next 25 • 19 categories Category difference 25 cases 100, 00% 0 1 2 3 4 5 6 7 8 9 7 6 5 4 0 2 0 1 0 0 28% 24% 20% 16% 0% 8% 0% 4% 0% 0% Directional accuracy: 14 Directional stability: 56%

Test on Crude Futures with Classical Open High Low Close Directional stability 41, 67% 50, 00% 54, 17% 45, 83% Correlation 0, 83 0, 85 0, 90 0, 85

Test on sales time series of one product • 100 weeks used to forecast the next 50 • 7 categories Category difference 0 1 2 3 4 5 6 50 cases 24 17 4 2 1 2 0 100, 00% 48% 34% 8% 4% 2% 4% 0% Directional accuracy: 33 Directional stability: 66%

Test on on sales time series of one product Benchmark • 100 weeks used to find the most common Quantity of cathegories 50 40 30 20 10 0 1 2 3 4 5 6 7

Test on sales time series of one product – Benchmark results • 100 weeks used to forecast the next 50 • 7 categories Category difference 0 1 2 3 4 5 6 50 cases 23 26 1 0 0 100, 00% 46% 52% 2% Directional accuracy: 30 Directional stability: 60% 0% 0%

Summary of tests • Crude Futures directional stability: – Similarity Searching Model: 72% – Benchmark: 68% • Product sales time series directional stability: – Similarity Searching Model: 66% – Benchmark: 60%

Thank you for your attention!