Beruflich Dokumente
Kultur Dokumente
Team: Jacob Epstein, Emre Kiziltug, Alexander Rubin | Project Lead: Emma Rasiel | Project Manager: Hang Su
*In descending order. Bonds issued in JPY not considered due to errors in data
○ Corporate Bond: Debt issued by a company to raise money
○ Arbitrage: The simultaneous buying and selling of nearly Graphic Analysis of Characteristics:
identical assets with differing prices to make riskless returns
○ Yield: Rate of return on a bond investment Maturities Peak at 5, 10, 30 years More Outstanding for Identified Bonds
○ Inversely correlated with price
○ Option-Adjusted Spread (OAS): Adjusted measure of yield
○ Duration: Price sensitivity to changing interest rates
Methods
Process:
Not Actionable
○ Iterate through pairs of bonds issued by same company, filter to Higher Quality Bonds More Likely to be Identified
ensure sufficient similarity
○ Same issuer, maturity and quality; high correlation
Key
○ Check for opportunities – 50 basis point (0.5%) difference in
duration- adjusted OAS
○ Visualize similar bonds over time
N
○ Classify with K-Nearest Neighbor supervised learning algorithm
20077
○ Compare common bond properties of identified bonds vs. 2090
baseline
Classification
Visualization of Filtering Technique:
K-Nearest Neighbor Classification: Conclusions
○ Need to identify the actionable cases from the 5000 identified bond pairs
○ Trained and validated on 300 manually classified observations Selected Conclusions from Analysis:
○ Chose k = 18 (square root of training set size) ○ Opportunities for short-term arbitrage exist regularly during
○ Engineered eight features non-crisis times
○ Proportion of points outside bounds, consecutive points crossing lower bound, upper
○ Most common among bonds that demonstrate the above
bound and median, same currency, correlation, volatility, squared difference from
properties, which indicates opportunities are most common in
median
liquid markets
○ Used min-max normalization to scale all features between 0 and 1
○ Identified 2090 good arbitrage cases ○ A viable trading strategy based on corporate bond arbitrage is
○ Evaluated with 10-fold cross validation possible, and is aided by the high liquidity of the identified bonds
○ Precision: 81.6%
○ Recall: 79.5%