Diversification in crypto is an illusion that most traders do not realize until a crash reveals it. You run five bots on five different altcoins — SOL, AVAX, DOGE, LINK, SHIB — and you believe your risk is spread across five independent assets. Then Bitcoin drops 10 percent in an hour and all five altcoins drop 12 to 18 percent simultaneously. Your diversified portfolio behaves like a single concentrated bet.
This happens because crypto assets are highly correlated, especially during stress events. The baseline cross-asset correlation among major altcoins is approximately 0.6 in normal conditions and approaches 0.9 during selloffs. Five altcoin bots with equal sizing create far more concentrated risk than five bots in genuinely uncorrelated markets.
The Correlation Problem in Practice
We run 45 paper trading bots across six strategy types. Thirteen of those bots trade mean reversion on Bollinger Bands across 13 different altcoin symbols. Five trade momentum on another set of altcoins. Six trade 4-hour momentum on BTC, ETH, SOL, ADA, SHIB, and AVAX. If all 24 of these bots independently decide to go long during a market euphoria phase, our portfolio has massive directional exposure to crypto despite appearing diversified across strategies and symbols.
This is not hypothetical. During euphoria phases, mean reversion strategies on altcoins naturally go long after post-euphoria dips, while momentum strategies go long because price is trending upward. Different strategies converge on the same direction when market conditions are extreme. The diversification that exists during normal conditions evaporates precisely when you need it most.
How Our Correlation Sizer Works
Every order passes through a correlation-aware position sizer that adjusts the trade size based on how correlated the new position would be with the existing portfolio. The adjustment range is 0.25 to 1.0. A multiplier of 1.0 means no reduction (the new position is uncorrelated with existing exposure). A multiplier of 0.25 means the position is cut to 25 percent of its intended size because the portfolio is already heavily exposed in the same direction.
The calculation works in two parts. First, it measures concentration: what fraction of existing exposure is in the same asset versus different assets. Same-asset exposure is weighted at 1.0 (perfect correlation with itself). Cross-asset exposure is weighted at the default cross-asset correlation of 0.6 (the empirical baseline for altcoin pairs).
Second, it applies a same-asset penalty. If you already have three or more positions in the same symbol (from different strategies), the penalty maxes out at 1.0. Fewer existing same-asset positions produce a proportionally smaller penalty. The threshold is three positions, reflecting the maximum number of concurrent strategies we run on any single symbol.
The raw concentration score combines these two components with 60 percent weight on the notional concentration and 40 percent weight on the same-asset penalty. The final multiplier is 1.0 minus 0.75 times the raw score, with a floor at 0.25. This means even in the worst case (heavily concentrated portfolio with multiple same-asset positions), the sizer never eliminates a trade entirely. It reduces to one quarter of the intended size, preserving the signal while containing the risk.
A Worked Example
Consider a bot that generates a long signal on AVAX/USDT with an intended position of 100 dollars. The current portfolio has the following open positions:
- AVAX/USDT long: 80 dollars (from a different strategy)
- SOL/USDT long: 90 dollars
- DOGE/USDT long: 70 dollars
- LINK/USDT long: 85 dollars
Total notional exposure is 325 dollars. Same-asset (AVAX) notional is 80 dollars. Cross-asset notional is 245 dollars. Same-asset weight is 80 divided by 325, which is 0.246. Cross-asset weight is 245 divided by 325, which is 0.754.
Concentration score from notional: (0.246 times 1.0) plus (0.754 times 0.6) equals 0.246 plus 0.452 equals 0.698. Same-asset penalty: 1 existing AVAX position divided by 3 (maximum), equals 0.333.
Raw score: (0.6 times 0.698) plus (0.4 times 0.333) equals 0.419 plus 0.133 equals 0.552. Final multiplier: 1.0 minus (0.75 times 0.552) equals 1.0 minus 0.414 equals 0.586.
The 100 dollar intended position becomes 58.60 dollars. The sizer reduces the position by 41 percent because the portfolio already has significant long exposure to correlated altcoins, including an existing AVAX position. This is a meaningful reduction that prevents the portfolio from becoming dangerously one-directional.
Portfolio-Level Caps as a Backstop
The correlation sizer is a continuous adjustment: every position gets sized according to the current portfolio composition. But continuous adjustments alone cannot prevent extreme scenarios where many signals fire simultaneously.
This is why the portfolio-level exposure cap at 50 percent of aggregate capital and the single-asset concentration cap at 25 percent serve as hard backstops. The correlation sizer gracefully reduces position sizes as concentration increases. The portfolio caps stop new positions entirely when concentration reaches dangerous levels.
In our 45-bot portfolio with 45,000 dollars aggregate capital, the 50 percent exposure cap means no more than 22,500 dollars in open positions total. The 25 percent single-asset cap means no more than 11,250 dollars on any single symbol. These are absolute limits that override the correlation sizer's output.
Why 0.6 Default Correlation
The 0.6 default cross-asset correlation for altcoins is an empirical estimate from our 30-day rolling correlation analysis. During normal market conditions, the average pairwise correlation among our 13 altcoin symbols (SHIB, DOGE, AVAX, SOL, LINK, SUI, PEPE, WIF, NEAR, ARB, OP, APT, INJ) ranges from 0.45 to 0.75. The 0.6 value is conservative: it assumes moderate correlation even when recent data might show lower values, because correlation increases during stress events exactly when you need the protection most.
A lower default (say 0.3) would underestimate the true correlation during market stress, leading to oversized positions that become dangerously correlated when BTC sells off. A higher default (say 0.8) would be too conservative during normal conditions, unnecessarily shrinking positions and reducing returns. The 0.6 value balances these concerns.
Our correlation regime strategy (one of our deployed macro strategies) directly measures BTC-altcoin correlation and adapts its trading behavior based on the current regime. When correlation is elevated, it reduces exposure. When correlation breaks down, it increases exposure. The correlation sizer applies the same principle at the portfolio construction level rather than at the individual strategy level.
Strategies That Provide Real Diversification
True diversification in our portfolio comes not from trading different altcoins (which are correlated) but from trading different market dimensions. Our six deployed strategy types operate on different signals:
Mean reversion trades price oscillations. Momentum trades price trends. These two are naturally somewhat anti-correlated because mean reversion performs better in ranging markets while momentum performs better in trending markets. Running both provides genuine strategy diversification.
The 4-hour momentum strategy on BTC and ETH provides asset diversification because major assets have different dynamics than altcoins. Leverage composite trades derivatives data (open interest, funding rates, long-short ratios), which is a completely different information source than price-based strategies. Correlation regime trades macro cross-asset patterns. NUPL cycle filter trades on-chain analytics. Stablecoin supply momentum trades capital flow data.
Each of these six strategy types is driven by different market phenomena. When mean reversion suffers during a trending market, momentum benefits. When price-based strategies are confused by choppy conditions, on-chain and macro strategies may still detect directional capital flows. This is diversification at the signal level, not just at the asset level, and it is far more robust than spreading the same strategy type across correlated altcoins.
The Sizing Hierarchy
Position sizing in our system works as a layered hierarchy. Risk parity allocates capital across bots based on inverse volatility. Half-Kelly determines the base risk per trade. The correlation sizer adjusts for portfolio overlap. Per-bot risk gates enforce hard limits (25 percent max position, 20 percent max drawdown, 5 percent daily loss). Portfolio-level caps prevent aggregate overexposure (50 percent total, 25 percent single asset).
Each layer addresses a different aspect of the sizing problem. No single formula or threshold can handle all of them. The correlation sizer specifically addresses the gap between individual-bot risk management and true portfolio diversification. Without it, 45 individually risk-managed bots can still create a portfolio that behaves like a single concentrated bet during market stress.