Trend Classifier

Start with two: 50-day for intermediate trend and 200-day for long-term trend. Price position relative to these plus their slopes gives a good basic picture. Add 20-day for shorter-term if needed.

ADX above 25 is generally considered a strong trend. Below 20 suggests weak or no trend (range-bound). Between 20-25 is developing. Remember ADX measures strength, not direction.

A swing high is a price bar with lower highs on both sides (local peak). A swing low has higher lows on both sides (local trough). Look for bars where price reversed direction. Many platforms can mark these automatically.

Conflicting signals often indicate transition or weak trend. When MA says uptrend but ADX is low, it's a weak uptrend. Give more weight to agreement. Conflicting signals = smaller position or wait.

For swing trading, update daily. Check at market close whether today's action changes the classification. Major changes are rare; most updates confirm existing classification.

Watch for: 1) Momentum divergences (price higher, RSI lower), 2) ADX declining from high levels, 3) First lower low in uptrend or higher high in downtrend, 4) Volume divergences. These are warnings, not confirmations.

Check weekly for overall bias, daily for trading signals, hourly for entry timing. Trade in the direction of the higher timeframe. If weekly is up, daily pullback is buying opportunity. If timeframes conflict, reduce size.

Early: Just crossed MA, ADX rising from low. Mature: Clear separation from MAs, ADX stable high. Late: Extended far from MAs, ADX declining from peak, divergences present. Each phase has different trading implications.

When ADX falls below 20, MAs flatten and intertwine, and price starts oscillating in a range. Don't switch on first day - wait for confirmation over 3-5 days. The transition period is trickiest.

Best trades: strong stock in strong sector in strong market. Check sector ETF trend (XFN for financials, XEG for energy). If sector is weak but stock is strong, it's relative strength but riskier. Align with sector when possible.

Use R/S analysis: divide series into subperiods, calculate range/standard deviation for each, regress log(R/S) on log(period length). Slope is Hurst exponent. Python: nolds.hurst_rs(). H > 0.5 = trending, H < 0.5 = mean-reverting.

Use hmmlearn (Python): define 2-3 states (up/sideways/down), fit to return series, use model.predict() for most likely state or model.predict_proba() for state probabilities. Specify emission as Gaussian. Validate with out-of-sample data.

Options: 1) Forward return sign (>1% = up, <-1% = down, else sideways), 2) Future price vs current (higher/lower/same), 3) Ex-post trend identification. Be careful with lookahead bias. Use proper time-series CV.

1) Define individual classifiers (rules, regression, ML, HMM). 2) Get classification from each. 3) Combine via voting (majority wins) or averaging (if numeric). 4) Confidence = agreement level. Track each model's contribution to performance.

Analyze historical trend durations. Fit survival model (exponential, Weibull). Calculate hazard rate: P(trend ends | has lasted T days). Can also use HMM transition matrix: expected duration = 1/(1-P(stay in state)). Use for position management.