Most crypto portfolio management advice amounts to picking the right coins and rebalancing periodically. This works for passive holders but is completely inadequate for active trading portfolios. When you run 45 automated bots across 6 strategy types and 13 symbols, portfolio management means capital allocation, correlation monitoring, risk budgeting, decay detection, and multi-level constraint enforcement.
Capital Allocation
Our current allocation is 1,000 dollars per bot across 45 bots, totaling 45,000 dollars. This equal allocation is deliberate for the paper trading phase: it simplifies monitoring and makes per-bot performance directly comparable.
For live trading, we will switch to risk parity allocation. Each bot receives capital inversely proportional to its volatility. Lower-volatility strategies get more capital because they contribute less risk per dollar allocated. Higher-volatility strategies get less capital because a small allocation already contributes significant risk.
The capital allocator enforces a 40 percent single-bot cap to prevent any one strategy from dominating the portfolio regardless of how low its volatility appears. Surplus capital from capped bots is redistributed to uncapped bots using the same inverse-volatility weights.
Half-Kelly sizing determines the per-trade risk within each bot. The Kelly fraction (win rate times payoff ratio minus loss rate, divided by payoff ratio) is multiplied by 0.5 for the conservative half-Kelly variant. This preserves roughly 75 percent of theoretical growth rate while cutting drawdown variance nearly in half.
Correlation Management
The biggest risk in a crypto portfolio is not individual strategy failure. It is correlated failure. Crypto altcoins have a baseline cross-asset correlation of approximately 0.6 in normal markets, rising to 0.9 during crashes. Running 13 mean reversion bots on 13 different altcoins feels diversified but behaves like a single concentrated bet when BTC drops.
Our correlation sizer adjusts position sizes based on portfolio overlap. The multiplier ranges from 0.25 (heavily correlated portfolio, reduce to quarter size) to 1.0 (uncorrelated, full size). The calculation weights same-asset exposure at 1.0, cross-asset exposure at the default 0.6 correlation, and applies a same-asset penalty when more than 3 positions exist in the same symbol.
Portfolio-level caps provide hard backstops. Total exposure cannot exceed 50 percent of aggregate capital. Single-asset concentration cannot exceed 25 percent. These are absolute limits that override individual bot sizing decisions.
Strategy Diversification
True diversification comes from different signal sources, not different assets. Our six deployed strategy types operate on fundamentally different market dimensions.
Mean reversion and momentum are naturally anti-correlated because they respond to opposite market conditions. Running both ensures that at least one strategy type aligns with current conditions at any time.
Derivatives strategies (leverage_composite) use open interest, funding rates, and long-short ratios — information invisible on a spot price chart. The signal source is structurally different from price-based strategies.
Cross-asset macro strategies (correlation_regime) monitor BTC-altcoin correlation dynamics. On-chain analytics strategies (nupl_cycle_filter, stablecoin_supply_momentum) use blockchain data (NUPL, stablecoin market cap) that is uncorrelated with any technical indicator.
The correlation between macro/on-chain strategy returns and price-based strategy returns is consistently below 0.20 in our backtests. This is genuine diversification at the signal level, not the illusion of diversification from spreading the same strategy across correlated assets.
Decay Monitoring
Strategies that worked last month may stop working this month. Market conditions shift, volatility regimes change, and edges can degrade. Our decay detector evaluates each bot's rolling 30-day Sharpe ratio from per-trade PnL. When the Sharpe drops below 0.5 (with at least 10 trades for statistical significance), the bot is automatically paused.
The 0.5 threshold is conservative. Our deployed strategies have validated Sharpe ratios from 1.7 to 19.0. A rolling Sharpe of 0.5 represents massive degradation from expected performance. The detector catches genuine strategy failure, not normal variance.
Beyond the automated detector, our daily monitoring routine (enforced by the dead man's switch's 24-hour check-in requirement) provides qualitative oversight. The drawdown chart, risk event log, and per-bot performance metrics are reviewed daily. Visual trends often reveal degradation before the statistical threshold triggers.
Risk Budgeting
Each strategy type receives an implicit risk budget based on its validated performance and the risk constraints it must operate within. Mean reversion bots have the largest aggregate allocation (13 bots, 13,000 dollars) because they have the highest validated Sharpe ratios and the lowest Monte Carlo drawdown profiles.
Macro and on-chain strategies have moderate allocations (18 bots, 18,000 dollars combined) with modest Sharpe expectations. Their value is diversification: the low correlation with price-based strategies improves the portfolio's risk-adjusted return even though their individual Sharpe ratios (0.11 to 1.45) are lower.
Derivatives strategies have the smallest allocation (3 bots, 3,000 dollars) because they have the shortest validation history (3 sub-periods instead of 5 due to limited Coinglass data). As more data accumulates and the validation strengthens, this allocation will increase.
The Portfolio Drawdown Halt
The most important portfolio-level mechanism is the 15 percent aggregate drawdown halt. If total portfolio equity declines 15 percent from its peak, all 45 bots stop. This is more conservative than the 20 percent per-bot limit because a portfolio-level drawdown indicates systemic stress, not individual bot failure.
The halt requires manual intervention to reset, ensuring a human reviews the situation. The questions to answer before resuming: Was the drawdown caused by a market event (and conditions have stabilized)? Was it caused by strategy decay (and the decay detector should have caught it earlier)? Was it caused by a correlated failure across strategies (suggesting the correlation assumptions need updating)?
This manual review is the portfolio management process working as intended. The automated systems handle normal operations. The human handles the exceptions that automated rules cannot anticipate.
Measuring Portfolio Performance
Portfolio performance is measured at the aggregate level using the same metrics applied to individual strategies: Sharpe ratio, maximum drawdown, Calmar ratio, and probability of profit from Monte Carlo simulation. The portfolio Sharpe should be higher than the weighted average of individual strategy Sharpes because diversification reduces variance.
We track portfolio equity hourly (idempotent snapshots per hour, both per-bot and aggregate). The 30-day drawdown chart and exposure breakdown charts on the risk dashboard provide real-time portfolio health monitoring. Strategy risk scorecards show per-strategy Sharpe, Sortino, Kelly fraction, and risk score from backtest aggregates.
The goal is a portfolio that produces consistent returns across market regimes with controlled drawdowns. Individual strategies may have volatile periods, but the portfolio as a whole should be smoother because the strategies respond to different market dimensions and naturally offset each other during regime transitions.