StrategiesResearchQuant

A Comparative Study of Quantitative Trading Strategies

April 15, 202612 min read

Artic ships with over 30 quantitative strategies spanning five families. Rather than picking one, an LLM evaluates market conditions and selects the optimal strategy — or blends several — in real time. This post breaks down each family, explains the core signal logic, and compares their strengths so you can understand what is happening under the hood.

Every strategy in Artic returns a normalised signal between −1 (strongly bearish) and +1 (strongly bullish), along with a detail string for diagnostics. The LLM reads these signals, cross-references volatility and regime data, and decides position sizing and direction.

Category	Best Market Fit	Risk	Signal Latency
Momentum	Strong trends	Medium-High	Low
Mean Reversion	Range-bound	Medium	Low
Volatility	Breakouts & squeezes	High	Medium
Volume / Order Flow	Intraday, liquid markets	Low-Medium	Low
Statistical	All regimes	Low	Medium-High

1. Momentum Strategies

Momentum strategies bet that assets moving strongly in one direction will continue. They are the bread and butter of trend-following systems and work best in directional markets with clear price trends.

Simple Momentum

% return over N bars

The simplest signal: compute percentage return over a lookback window. Positive return → bullish, negative → bearish. Fast to compute, but prone to whipsaws in choppy markets.

Dual Momentum

short-term vs long-term divergence

Compares short-term momentum against long-term momentum. When both agree, the signal is strong. When they diverge, it can detect regime shifts early — at the cost of occasional false starts.

Breakout

N-period high/low break

Fires a full ±1.0 signal when price breaks the highest high or lowest low over N periods. Binary and decisive — ideal for catching strong moves, but generates false signals in sideways markets.

Donchian Channel

normalised position in channel

Maps price position within the Donchian Channel (highest high to lowest low). Near the top → bullish, near bottom → bearish. Smoother than raw breakout, good for position sizing.

MA Crossover

(fast SMA − slow SMA) / slow

Classic moving average crossover. Signal magnitude scales with the gap between fast and slow SMA, giving proportional conviction rather than binary entries.

EMA Crossover

exponential MA cross

Same logic as MA Crossover but uses exponential averages. Reacts faster to recent price changes, which helps in volatile crypto markets but increases whipsaw risk.

MACD Signal

MACD line vs signal line

Measures the divergence between the MACD line and its signal line. Combines trend detection with momentum strength — widely used across traditional and crypto markets alike.

Ichimoku

Tenkan/Kijun cross + cloud position

Multi-timeframe system: Tenkan-sen / Kijun-sen crossover layered with price position relative to the Kumo cloud. Rich signal but requires more data for warm-up.

Momentum strategies dominate in trending markets. In Artic, the LLM up-weights them when ADX (trend strength) is high and volatility is expanding.

2. Mean Reversion Strategies

Mean reversion strategies assume that prices oscillate around a fair value and will revert after extreme moves. They work best in range-bound, liquid markets and tend to underperform during strong trends.

Z-Score

inverted deviation from rolling mean

Computes how many standard deviations price sits from its rolling mean, then inverts it. A z-score of +2 generates a bearish signal (expecting reversion down). Simple, robust, and parameters-light.

Bollinger Reversion

fade Bollinger Band extremes

When price touches or exceeds the upper Bollinger Band, go short; when it hits the lower band, go long. Signal intensity scales with how far price has penetrated the band.

RSI Signal

oversold (<30) / overbought (>70)

The classic Relative Strength Index. Below 30 → long, above 70 → short. Signal is proportional to the distance from the 50 midline. Works well with confirmation from other signals.

Stochastic Signal

%K/%D extremes

Similar logic to RSI but uses the Stochastic Oscillator (%K and %D lines). More sensitive to short-term moves, making it useful for intraday entries.

Range Support/Resistance

fade recent highs/lows

Identifies recent local highs and lows as support/resistance levels, then fades moves into those zones. Simple but effective in markets that respect range boundaries.

Linear Regression Channel

deviation from trend line

Fits a linear regression to price, then measures deviation from the trend line. Combines mean reversion with trend awareness — revert to the trend, not just the mean.

Mean reversion strategies are the counterbalance to momentum. The LLM switches to them when ADX is low and Bollinger Bandwidth is contracting, indicating a range-bound regime.

3. Volatility Strategies

Volatility strategies don't bet on direction per se — they bet on the expansion or contraction of price movement. They are especially valuable around events, news releases, and regime transitions.

ATR Breakout

candle range > ATR × multiplier

Fires when a single candle's range exceeds a multiple of the Average True Range. Catches explosive moves early but needs tight risk management as the trade is already extended.

Bollinger Squeeze

low-vol consolidation → breakout

Detects when Bollinger Bands narrow (squeeze), indicating compressed volatility, then signals the direction of the breakout. High win rate on the initial expansion but requires patience during the squeeze phase.

Keltner-Bollinger

Keltner bands vs Bollinger bands

Compares ATR-based Keltner Channels against standard-deviation-based Bollinger Bands. When Bollinger fits inside Keltner, a squeeze is on. Breakout direction determines signal. More nuanced than Bollinger Squeeze alone.

Volatility strategies act as regime detectors. Artic's LLM uses them as confirming filters — if volatility is compressing, it reduces position size across all strategies until a breakout confirms direction.

4. Volume & Order Flow Strategies

Price tells you what happened; volume tells you how much conviction was behind it. These strategies incorporate trade volume and derivatives data to gauge the quality of moves.

VWAP Deviation

distance from volume-weighted avg price

VWAP is the benchmark for institutional execution. When price deviates significantly above VWAP, short-term traders tend to sell; below, they buy. Artic uses this as an intraday anchor.

OBV Trend

On-Balance Volume slope

Cumulates volume on up-days and subtracts volume on down-days. Rising OBV confirms bullish moves; divergence between OBV and price is a powerful early warning for reversals.

Funding Bias

fade extreme funding rates

Specific to perpetual futures. When funding rate is extremely positive (longs paying shorts), the market is over-leveraged long — and vice versa. Fading extremes captures reversion in derivatives markets.

Volume strategies add a conviction layer. The LLM uses OBV divergence to validate or override momentum signals — a rising price with falling OBV triggers caution.

5. Statistical & Advanced Strategies

These strategies use more sophisticated mathematical models. They tend to be more robust to noise but require longer warm-up periods and more data to produce stable signals.

Kalman Fair Value

1D Kalman filter price estimate

Uses a Kalman filter to estimate the 'true' price, filtering out noise. Trades deviations from this fair value estimate. Adapts its smoothing dynamically based on recent forecast accuracy.

ADX Filter

+DI / −DI with trend strength

The Average Directional Index measures trend strength, not direction. Artic uses it as a meta-signal: high ADX → enable momentum strategies; low ADX → switch to mean reversion.

Supertrend

ATR-based dynamic support/resistance

Plots dynamic stop levels above and below price using ATR. When price crosses above the stop, it flips bullish (and vice versa). Clean binary signals with built-in risk levels.

Statistical strategies serve as the backbone of Artic's meta-layer. ADX and Kalman outputs directly influence how the LLM weights every other strategy category.

How the LLM Selects Strategies

The core innovation in Artic is not any individual strategy — it is the dynamic selection layer. Here is how it works:

Regime Detection

ADX, Bollinger Bandwidth, and volatility metrics classify the current market as trending, ranging, or transitioning.

Signal Collection

All 30+ strategies compute their signals independently. Each returns a normalised value and diagnostic detail.

LLM Evaluation

The language model receives signal values, regime classification, recent performance data, and risk constraints. It reasons about which strategy family fits the current conditions.

Position Sizing

Based on conviction level and volatility, the LLM sets position size. High-confidence regime matches get larger allocations; ambiguous conditions get scaled down.

Continuous Adaptation

Every tick cycle, the LLM re-evaluates. If the regime shifts mid-trade, it can reduce exposure or rotate to a different strategy — no manual intervention needed.

Conclusion

No single strategy wins in all market conditions. Momentum thrives in trends but bleeds in ranges. Mean reversion captures bounces but gets steamrolled by breakouts. Volatility strategies detect regime shifts but need patience.

The insight behind Artic is that the selection problem — which strategy to use right now — is itself a reasoning problem. Language models excel at synthesising disparate signals, weighing trade-offs, and making contextual decisions. By pairing 30+ deterministic quant strategies with an LLM orchestrator, Artic gets the best of both worlds: rigorous, tested signal generation and flexible, adaptive strategy selection.

All posts