Robo-Advisory & WealthTech

Modern Portfolio Theory (Markowitz): 1) Collect expected returns and standard deviation (volatility) for each asset class, 2) Build covariance/correlation matrix between assets, 3) Run mean-variance optimization — find the portfolio on the Efficient Frontier maximizing return for a given risk level, 4) The Sharpe Ratio = (Return - RiskFreeRate) / StdDev is maximized, 5) Black-Litterman extends this by incorporating investor views into expected returns. In practice, robo-advisors map investor risk scores to pre-defined model portfolios (Conservative = 20% equity / 80% debt; Aggressive = 80% equity / 20% debt) rather than re-running full optimization per user.

XIRR (Extended Internal Rate of Return) is used for investments with irregular cash flows (like SIPs). CAGR works for lump-sum, but SIPs have multiple purchase dates at different amounts. XIRR finds the single discount rate r such that NPV of all cash flows = 0: Sum[ Amount_i / (1+r)^(date_i/365) ] = Current_Value. Example: ₹5,000/month for 3 years, current value ₹2.2L — XIRR ~12%. This is the true annualized return accounting for the timing of each installment. XIRR > CAGR when recent performance is good (more units purchased earlier when NAV was low).

Tax-Loss Harvesting: Sell losing funds to realize capital losses, which offset capital gains, reducing tax liability. Example: Equity fund A has ₹50K LTCG, Equity fund B has ₹20K unrealized loss — sell B to offset, tax only on ₹30K gain. In India: LTCG on equity >₹1 lakh taxed at 10%, STCG at 15%. Limitations: 1) Wash-sale rules — India doesn't have formal wash-sale rules like the US, but re-purchasing the same fund too quickly may trigger tax scrutiny, 2) Exit loads (1% if redeemed within 1 year), 3) Only worthwhile if LTCG exceeds ₹1L free limit, 4) Complexity increases with many SIP installments having different holding periods.

Challenges: All SIPs due on 5th of month create huge spike. Solution: 1) Batch processing — group SIPs by fund and RTA, send bulk orders to CAMS/KFintech, 2) NACH auto-debit — pre-scheduled with NPCI, bank debit happens automatically, 3) Queue-based architecture — Kafka topic with partitions per fund house, consumers process in parallel, 4) Idempotency — each SIP execution has idempotency key to prevent duplicate orders, 5) Retry with backoff — failed debits go to retry queue, 6) RTA limits — throttle to CAMS/KFintech API rate limits, 7) Spread execution dates — encourage investors to choose different dates (5th, 10th, 15th), 8) Async processing — execute in background, push notifications on completion.

Traditional: Single portfolio, maximize returns vs market benchmark (Nifty 50). Goal-based: Each goal (retirement, education, house) has its own sub-portfolio with specific target, time horizon, and risk profile. Key differences: 1) Allocation varies per goal — house down payment in 2 years → debt funds; retirement in 25 years → aggressive equity, 2) Progress tracking — 'You are 65% of the way to your retirement goal', 3) SIP sizing — reverse-engineered from goal amount and expected returns, 4) Rebalancing is goal-specific, 5) Investor behavior — goal context reduces panic selling (attached to outcome, not daily NAV). Systems challenge: Managing multiple sub-portfolios per investor, attributing performance per goal, separate rebalancing logic per goal.