Portfolio Management — Position Sizing & Risk Limits

A single brilliant trade means nothing if poor portfolio management turns a winning system into a losing one. Professional prediction market traders focus as much on position sizing and risk controls as on signal generation. This lesson covers the portfolio management layer that sits above your trading strategies and governs how your Oracle allocates capital.

The Three Pillars of Prediction Market Portfolio Management

Pillar 1 — Position sizing: How much capital to allocate per trade. Too much and a single bad trade wipes you out. Too little and your returns are meaningless.

Pillar 2 — Exposure limits: Maximum capital in any single market, market category, or strategy at any time.

Pillar 3 — Drawdown controls: What to do when your bankroll decreases below certain thresholds. Automatic stop mechanisms that prevent a losing streak from cascading into a catastrophic loss.

Position Sizing: The Half-Kelly Approach

The Kelly Criterion provides the mathematically optimal fraction of bankroll to bet for maximum long-run growth rate. For prediction markets with known probabilities:

Kelly fraction = (p × b - q) / b

Where:

p = probability of winning (your ensemble estimate)
q = 1 - p
b = net odds (payout per $1 risked = (1 - current_price) / current_price)

Full Kelly is mathematically optimal but psychologically difficult — it can recommend 30%+ of bankroll per trade, leading to extreme volatility. Half-Kelly is the professional standard: use half the mathematically optimal fraction.

Practical implementation with hard limits:

Maximum per trade: 5% of bankroll (Theta Sniper), 3% (Geo Scout), 2% (new unproven strategies)
Minimum trade size: $10 / PKR 2,800 (below this, transaction costs are proportionally too large)
Total exposure cap: 40% of bankroll deployed at any time (keeping 60% dry powder)

Exposure Diversification

Don't concentrate all positions in one category. Example exposure limits:

Category	Max Portfolio Allocation
SBP/Pakistan macro	15%
Cricket markets	10%
Global geopolitics	20%
US politics/policy	15%
Other markets	20%
Cash (undeployed)	20% minimum

These limits force diversification. Even if your Pakistan macro edge is very high, concentrating 50% of capital in SBP trades means you're making a macro bet on Pakistan rather than running a diversified prediction market portfolio.

The Drawdown Control System

python

class PortfolioManager:
    def __init__(self, initial_bankroll: float):
        self.peak_bankroll = initial_bankroll
        self.current_bankroll = initial_bankroll
        self.SOFT_STOP_PCT = 0.85   # 15% drawdown: reduce position sizes
        self.HARD_STOP_PCT = 0.75   # 25% drawdown: stop all new trades

    def update_bankroll(self, new_value: float):
        self.current_bankroll = new_value
        if new_value > self.peak_bankroll:
            self.peak_bankroll = new_value

    @property
    def drawdown_pct(self) -> float:
        return self.current_bankroll / self.peak_bankroll

    @property
    def trading_status(self) -> str:
        dd = self.drawdown_pct
        if dd >= 1.0:
            return "FULL"      # Normal operations
        elif dd >= self.SOFT_STOP_PCT:
            return "REDUCED"   # Cut position sizes by 50%
        else:
            return "HALTED"    # No new positions until recovery

    def get_position_size(self, base_size: float) -> float:
        status = self.trading_status
        if status == "FULL":
            return base_size
        elif status == "REDUCED":
            return base_size * 0.5
        else:
            return 0.0  # No new trades

Expected Monthly Performance Model

Based on realistic assumptions for a Pakistani Oracle bot:

Active markets monitored: 100-200 daily
Trades passing all filters: 3-7 per week
Average edge per trade: 12-18 cents
Average position size: $50-100 / PKR 14,000–28,000 (on $1,000 / PKR 280,000 bankroll at 5-10%)
Win rate with ensemble: 58-65%

Expected monthly P&L calculation:

20 trades/month × $75 average position × 15% average edge = $225 expected gross profit
Minus spread costs: $10-15
Minus API costs: $20-40
Net monthly expected profit: $170-200 on $1,000 bankroll = 17-20% monthly

At PKR exchange rates (approximately PKR 280/USD), that's PKR 47,600-56,000/month on a PKR 280,000 starting bankroll. These are expectations, not guarantees — actual results will vary significantly in the early months as you calibrate.

The Compounding Growth Plan

The most powerful portfolio management decision is reinvestment:

Month 1: Start with $500 (PKR 140,000). Target: $85-100 profit
Month 3: Reinvest all profits. Bankroll now $600-650. Target: $102-130/month
Month 6: Bankroll $750-900 if compounding. Monthly target: $127-180
Month 12: Bankroll $1,200-1,500 if consistently profitable

The key discipline: don't withdraw profits in the first 6 months. Compound everything back into the bankroll. A $500 starting position that compounds at 17%/month becomes $3,000 in 12 months — but only if you resist the temptation to cash out early.

Pakistan Case Study: The Drawdown That Would Have Wiped the Bankroll

Farhan from Gulberg Lahore had a good first 6 weeks. He started with $400 (PKR 112,000) and grew to $520. Confident in his system, he removed the 40% deployment cap from his bot — "my win rate is 68%, I can afford more exposure," he thought.

In week 7, three positions went against him simultaneously:

SBP rate market: unexpected political statement → loss of $28
Cricket market: rain interruption changed outcomes → loss of $19
Pakistan elections market: result closer than expected → loss of $31

Three consecutive losses totaling $78 — a 15% drawdown in one week.

Without the drawdown control system, his bot was still placing new trades throughout the losing week. By Friday, his bankroll was $412 — nearly back to his starting point after 6 profitable weeks.

With the PortfolioManager in place, what would have happened:

After loss #1 (-$28): Bankroll $492. Drawdown: 5.4%. Status: FULL — normal operations. After loss #2 (-$19): Bankroll $473. Drawdown: 9%. Status: FULL. After loss #3 (-$31): Bankroll $442. Drawdown: 15.0%. Status: SOFT STOP (REDUCED mode, position sizes halved).

All subsequent trades that week would have been at 50% normal size. If those also lost, the losses would be half as large.

Actual result vs. protected result:

Scenario	End of Week 7 Bankroll	Weeks to Recover
No drawdown controls	$412	3-4 weeks
With PortfolioManager	$435 (estimated)	1-2 weeks
With hard stop at 25%:	Trading halted at $390	Until manual review

The difference seems small — but it's the compounding effect that matters. Starting week 8 with $412 vs. $435 means his compound growth projections are permanently set back.

Farhan's lesson: "I thought drawdown controls were for beginners. They're for everyone. My 68% win rate meant nothing during a losing streak — the only thing that mattered was position sizing."

12-Month Bankroll Projection (PKR, starting PKR 112,000 / ~$400)

code

CONSERVATIVE SCENARIO (10%/month net return):
Month  1: PKR 123,200
Month  3: PKR 149,000
Month  6: PKR 198,000
Month 12: PKR 350,000

BASE CASE (17%/month net return):
Month  1: PKR 131,000
Month  3: PKR 179,000
Month  6: PKR 281,000
Month 12: PKR 784,000

UNDERPERFORMANCE (5%/month):
Month  1: PKR 117,600
Month  3: PKR 129,600
Month  6: PKR 142,800
Month 12: PKR 201,600

WARNING: These are projections, not guarantees.
Prediction markets involve real financial risk.
Start with PKR 30,000-60,000 ($100-200) and prove
your system over 3 months before scaling.

The real value is the SYSTEM — bot code, ensemble tuning,
database analytics, and discipline. Not the bankroll size.

Portfolio State Machine