Bullish EMA stack with rising EMA21 = long; scaled by ADX
Wilder (1978) ADX; Moskowitz, Ooi & Pedersen (2012) time-series momentumHow a call is produced
Every run pulls live WEEX perpetual candles, order book, trades, funding and open interest, builds an event dataset, runs 19 algorithms across 5 dimensions, walk-forward tests each one, fuses them by reliability and outputs a long or short call with reasons and a trade plan.
- 01
Data
- WEEX candles (taker-buy volume)
- Order book, 200 levels
- Last 1,000 trades
- Funding · mark/index
- Open interest · risk tiers
- 02
Features
- 49 features per candle
- Label: close after window > now
- Higher-timeframe resample
- Fractal swing points
- Estimated liquidation map
- 03
19 algorithms · 5 dims
- Trend & structure ×5
- Momentum ×3
- Volume & flow ×3
- Statistical ×4
- Live microstructure ×4
- 04
Walk-forward test
- Each candle scored with past data only
- ~300 verifiable samples
- Bayesian-shrunk hit rate
- Reliability 0.25×–1.8×
- 05
Fusion engine
- Weight = dimension budget × reliability
- S = Σ wᵢ·sᵢ
- Platt-calibrated P(up)
- Confidence 0–100
- 06
Decision
- LONG / SHORT + strength
- Reasons and counter-risks
- Entry / stop / targets
- Size and leverage cap
Fusion engine formulas
- Algorithm output
sᵢ ∈ [−1, 1] (direction × strength), cᵢ = |sᵢ| × historical credibility (live factors × 0.85)- Hit rate (shrunk)
hᵢ = (hits + 5) / (calls + 10), counting only samples with |sᵢ| ≥ 0.1- Reliability
rᵢ = clamp(1 + 12·(hᵢ − 0.5), 0.25, 1.8); live factors rᵢ = 1, missing data rᵢ = 0- Fusion weight
wᵢ = B(dimension) × rᵢ / Σ r(same dimension); B: trend 26%, momentum 14%, flow 20%, statistical 16%, micro 24%; missing dimensions are redistributed- Long/short score
S = Σ wᵢ·sᵢ; S > 0 long, S < 0 short; |S| ≥ 0.30 strong, ≥ 0.15 medium, else weak; |S| < 0.04 wait- Up probability
P(up) = σ(a + b·S), a and b Platt-fitted on the same fusion replayed over history (b limited to 1.5–6)- Confidence
100 × [0.38·min(1,|S|/0.35) + 0.27·agreement + 0.17·fused-backtest edge + 0.18·data completeness] × regime factor- Liquidation price
Long = entry × (1 − 1/L + mmr), short = entry × (1 + 1/L − mmr), L ∈ {5,10,20,50,100}; levels already traded are removed- Trade plan
Stop beyond the nearest swing / 1.2× expected move / order-book wall; targets 1.5R and 2.5R, snapped to a liquidation cluster or wall between 1R and 3R
The 19 algorithms and their sources
Trend & structureBudget 26%
Price above the SuperTrend line = long; stronger with distance
Seban (2009) SuperTrend; Wilder (1978) ATRPrice above cloud, Tenkan > Kijun, Chikou above price 26 bars ago = long
Hosoda (1969) Ichimoku Kinko HyoHigher highs + higher lows = long; close above last swing high (BOS) = strong long
Dow theory; Wyckoff; Williams (1995) fractalsClose above the 20-bar channel = long, below = short
Donchian (1960); Faith (2007) Turtle rulesMomentumBudget 14%
RSI > 50, rising positive MACD histogram, positive ROC = long
Wilder (1978) RSI; Appel (1979) MACD; Jegadeesh & Titman (1993)Fade moves beyond 1.2σ from SMA20 or stochastic extremes
Bollinger (2001); Lane (1950s) Stochastic; Poterba & Summers (1988)Bullish engulfing / hammer / three advancing = long; bearish counterparts = short
Nison (1991) Japanese Candlestick Charting TechniquesVolume & order flowBudget 20%
Rising OBV, positive CMF, high-volume up candles = long
Granville (1963) OBV; Chaikin Money FlowTaker buying dominance and rising CVD = long; price up while CVD falls = bearish divergence
Easley, López de Prado & O’Hara (2012) order-flow toxicity; Cont, Kukanov & Stoikov (2014)Above a rising VWAP = long; beyond 2.5 ATR it turns to reversion
Berkowitz, Logue & Noser (1988) VWAPStatistical learningBudget 16%
Trend and momentum of the 4× higher timeframe point up = long
Elder (1985) Triple ScreenHistorical up-frequency after the same 3-candle up/down state
Markov (1906); Fama (1965) random-walk testsFind the most similar historical windows and count what happened next
Cover & Hart (1967) nearest-neighbour classificationPredict up-probability from standardised features, walk-forward then refit
Cox (1958); Platt (1999) calibrationLive microstructureBudget 24% · live data, not backtested
More bid than ask depth near the price = long; large walls act as support/resistance
Cont, Kukanov & Stoikov (2014); Cao, Hansch & Wang (2009)Aggressive buy notional and large buys dominate = long
Lee & Ready (1991) trade classificationStrongly positive funding = crowded longs (contrarian short); negative = crowded shorts (long)
He, Manela et al. (2022) perpetual futures fundingDenser short liquidations above than long liquidations below = upward magnet
Estimated from volume profile, open interest and exchange risk tiersReferences
- Wilder, J. W. (1978). New Concepts in Technical Trading Systems. Trend Research.
- Appel, G. (1979). The Moving Average Convergence-Divergence Method.
- Bollinger, J. (2001). Bollinger on Bollinger Bands. McGraw-Hill.
- Granville, J. (1963). Granville’s New Key to Stock Market Profits.
- Hosoda, G. (1969). Ichimoku Kinko Hyo.
- Nison, S. (1991). Japanese Candlestick Charting Techniques.
- Donchian, R. (1960). High finance in copper. Financial Analysts Journal.
- Elder, A. (1985). Triple Screen Trading System.
- Jegadeesh, N. & Titman, S. (1993). Returns to buying winners and selling losers. Journal of Finance.
- Moskowitz, T., Ooi, Y. H. & Pedersen, L. H. (2012). Time series momentum. Journal of Financial Economics.
- Poterba, J. & Summers, L. (1988). Mean reversion in stock prices. Journal of Financial Economics.
- Berkowitz, S., Logue, D. & Noser, E. (1988). The total cost of transactions on the NYSE. Journal of Finance.
- Lee, C. & Ready, M. (1991). Inferring trade direction from intraday data. Journal of Finance.
- Cont, R., Kukanov, A. & Stoikov, S. (2014). The price impact of order book events. Journal of Financial Econometrics.
- Easley, D., López de Prado, M. & O’Hara, M. (2012). Flow toxicity and liquidity in a high-frequency world. Review of Financial Studies.
- Cover, T. & Hart, P. (1967). Nearest neighbor pattern classification. IEEE Transactions on Information Theory.
- Cox, D. R. (1958). The regression analysis of binary sequences. JRSS Series B.
- Platt, J. (1999). Probabilistic outputs for support vector machines. Advances in Large Margin Classifiers.
- He, S., Manela, A., Ross, O. & von Wachter, V. (2022). Fundamentals of perpetual futures. arXiv:2212.06888.
- Pardo, R. (2008). The Evaluation and Optimization of Trading Strategies (walk-forward analysis). Wiley.
Note: short-term crypto prices are noisy, and single technical algorithms usually hit 50–56% historically. The fusion engine weights many weak signals by measured reliability and adds live order-book, funding and liquidation information to give a direction and risk levels. Treat the call and plan as research, and manage your own size and risk.