Combining Strategies

How we combine trading strategies into trading subsystems

From strategy signals to forecasts

Each of the strategies we've discussed generates raw signals that need to be standardized before they can be combined (see those fancy formulas we've described in the Core Strategies section). To do this we convert each signal into a forecast that represents HyperCroc's conviction about future price direction on a standardized scale.

The signal transformation process

Raw signals vary wildly in magnitude and meaning. A momentum signal of 0.5 might be strong for one asset but weak for another. A breakout signal of 0.3 could be significant in low-volatility periods but trivial during high volatility. We need a common language.

We transform each raw signal into a forecast value that ranges from -1 (strong short conviction) to +1 (strong long conviction), with 0 representing no position. This standardisation serves multiple purposes: it makes signals comparable across different strategy types, allows for systematic combination rules, and provides clear position sizing guidance.

The transformation uses a scaling function calibrated to each strategy's historical distribution:

$$Forecast = clip[raw\_signal × scaling\_factor, -1, +1]$$

The scaling factor for each strategy is determined through historical analysis.

Diversification

Why combine multiple signals rather than pick the "best" one? Because different strategies capture different market phenomena, and combining them reduces the risk of any single strategy failing catastrophically.

Consider our momentum and breakout strategies. Momentum performs well in sustained trends but suffers during consolidation. Breakout strategies excel when volatility expands but underperform in steady trending markets. Carry strategies provide steady returns in range-bound markets but can face sharp losses during trend reversals. By combining these, we create a portfolio of strategies that's more robust than any individual approach.

The correlation between our strategy types shows why diversification matters. Check out the following figure, momentum strategies may act very similar, yet they capture different momentum effects:

Signal combination rules

For each asset we trade, we generate multiple signals from our different strategy variations. These need to be combined into a single forecast number that represents our overall conviction for that instrument.

Equal weights vs optimized weights

The simplest approach assigns equal weight to all signals. If we have 2 momentum variations (5-30d, 10-60d) and three breakout variations (10d,30d,60d) generating forecasts for ETH, we simply average them. This naive diversification works surprisingly well and avoids overfitting to past performance.

However, we can do better by considering each strategy's risk-adjusted performance and correlation structure. Our optimization seeks weights that maximize the Sharpe ratio of the combined forecast while respecting diversification constraints:

$$Combined\_Forecast = Σ(Weight_i × Forecast_i)$$

Subject to constraints that prevent excessive concentration in any single strategy and maintain stability across different market regimes. [TBD: Historical analysis to determine optimal forecast weights with rolling window optimization]

Volatility targeting: the core risk control

Rather than using fixed position sizes, we scale positions inversely to volatility to maintain consistent risk exposure.

The volatility target

We define our target volatility as the annualized standard deviation of daily portfolio returns we're willing to accept. This becomes the anchor for all position sizing decisions. Different portfolio profiles target different volatility levels, for example:

• Shallow Waters: 5% annual volatility target

• Swamp Wader: 15% annual volatility target

• Deep Swamp: 25% annual volatility target

• Apex Predator: Custom volatility targets

Position scaling calculation

For each instrument, we calculate the position multiplier that achieves our volatility target for a given trading instrument i:

$$Position\_Multiplier_i = (Target\_Volatility / \sigma_i) × (Portfolio\_Value / P_i)$$

This means when volatility doubles, our position size halves, maintaining constant risk exposure. For example, during the COVID crash when ETH volatility spiked to 200% annualized, our positions automatically scaled down by half. Conversely, during the low-volatility summer of 2023, position multiplier scaled up to maintain returns.