Six strategy patterns that actually work

There are infinite trading strategies. Most of them are variations on six core patterns. This guide walks through each — when it works, when it fails, and a code skeleton you can paste into WatchDog Bot and adapt.

01Momentum / Trend following

The bet: when prices have been moving in one direction, they'll keep moving in that direction long enough for you to capture some of it. Buy strength, sell weakness.

Classic implementations: moving average crossover, breakout from N-day high, RSI > 70 / < 30 signals.

Skeleton

import time
from collections import deque
import wd, ccxt

conn = wd.connection("Binance")
ex = ccxt.binance({"apiKey": conn.api_key, "secret": conn.api_secret, "enableRateLimit": True})

short_window, long_window = 12, 26
short = deque(maxlen=short_window)
long  = deque(maxlen=long_window)
position = 0

while True:
    try:
        price = float(ex.fetch_ticker("BTC/USDT")["last"])
        short.append(price); long.append(price)
        if len(long) < long_window:
            time.sleep(300); continue

        s, l = sum(short)/len(short), sum(long)/len(long)
        wd.log.info("price=%.2f short=%.2f long=%.2f position=%d", price, s, l, position)

        if s > l * 1.001 and position == 0:
            wd.log.info("BUY signal")
            if not wd.is_demo(): ex.create_market_buy_order("BTC/USDT", 0.001)
            position = 1
        elif s < l * 0.999 and position == 1:
            wd.log.info("SELL signal")
            if not wd.is_demo(): ex.create_market_sell_order("BTC/USDT", 0.001)
            position = 0

        time.sleep(300)
    except Exception as e:
        wd.log.error("tick failed: %s", e); time.sleep(60)

Key parameters to tune: the two MA windows, the crossover threshold (1.001 above to reduce noise), the polling interval, and the position size.

02Mean reversion

The bet: opposite of momentum — when prices stretch far from their recent average, they tend to snap back. Buy weakness, sell strength.

Classic implementations: Bollinger Band re-entry, z-score of price relative to N-period mean, pairs trading on correlated assets.

Skeleton (z-score reversion)

import time
import numpy as np
from collections import deque
import wd, ccxt

conn = wd.connection("Coinbase")
ex = ccxt.coinbase({"apiKey": conn.api_key, "secret": conn.api_secret})

window = 60
prices = deque(maxlen=window)
position = 0  # +1 long, -1 short, 0 flat
Z_ENTER, Z_EXIT = 2.0, 0.5

while True:
    try:
        p = float(ex.fetch_ticker("ETH/USD")["last"])
        prices.append(p)
        if len(prices) < window:
            time.sleep(60); continue

        mean = np.mean(prices); std = np.std(prices)
        z = (p - mean) / std if std > 0 else 0
        wd.log.info("p=%.2f mean=%.2f z=%+.2f pos=%+d", p, mean, z, position)

        # Enter when z is extreme; exit when it returns to mean
        if z < -Z_ENTER and position == 0:
            wd.log.info("BUY — price %.1f std below mean", abs(z))
            position = 1
        elif z > Z_ENTER and position == 0:
            wd.log.info("SELL — price %.1f std above mean", z)
            position = -1
        elif abs(z) < Z_EXIT and position != 0:
            wd.log.info("EXIT — z returned to %.2f", z)
            position = 0

        time.sleep(60)
    except Exception as e:
        wd.log.error("tick failed: %s", e); time.sleep(60)

03Market making

The bet: not on price direction at all. You profit by being the counterparty to impatient traders, earning the bid-ask spread on every round trip.

Classic implementations: simple quote-skew (covered in our Kalshi MM guide), Avellaneda-Stoikov, more complex inventory models.

Full implementation with inventory skew + adverse selection defenses: Kalshi Market Making Strategies.

04Arbitrage

The bet: the same asset trades at different prices in two places. Buy the cheap one, sell the expensive one, lock in the spread. Risk-free in theory; latency-bound in practice.

Classic implementations: cross-exchange spot arbitrage, triangular arb across three pairs, basis trading (spot vs futures), funding-rate harvesting.

Skeleton (spread scanner)

import time
import wd, ccxt

binance  = ccxt.binance({"enableRateLimit": True})
coinbase = ccxt.coinbase({"enableRateLimit": True})

THRESHOLD_BPS = 30
while True:
    try:
        a = float(binance.fetch_ticker("BTC/USDT")["last"])
        b = float(coinbase.fetch_ticker("BTC/USD")["last"])
        spread = abs(a - b) / min(a, b) * 10000
        wd.log.info("Binance %.2f  Coinbase %.2f  spread %.1f bps", a, b, spread)
        if spread > THRESHOLD_BPS:
            wd.log.warning("ARB OPPORTUNITY — would buy %s, sell %s",
                           "Binance" if a < b else "Coinbase",
                           "Coinbase" if a < b else "Binance")
            # Execution: buy on lower, sell on higher.
            # CAUTION: account for fees, slippage, and latency before going live.
        time.sleep(5)
    except Exception as e:
        wd.log.error(e); time.sleep(15)

05Dollar-cost averaging (DCA)

The bet: you don't try to time the market. You buy a fixed dollar amount on a schedule and let cost averaging do the work. Boring. Often beats active traders.

Full skeleton in the examples library.

06News / event-driven

The bet: markets react to news. If you can read news faster than humans (or interpret it better with an LLM), you can take positions before the price moves.

Classic implementations: headline scanners with LLM classifiers, sentiment scoring, economic calendar bots, social media signal extraction.

Full skeleton (with LLM classifier) in the examples library.

Choosing a pattern

A rough decision tree:

• You want to start simple and learn: DCA. Boring, useful, hard to mess up.
• You believe in the long-term direction of an asset: DCA + occasional momentum entry when conviction is high.
• You have an information edge (news, signals, expertise): News-driven, but be honest about latency.
• You like math and have patience: Market making — especially on Kalshi where retail edge exists.
• You enjoy chasing micro-edges: Arbitrage — be ready to invest in infrastructure.
• You're somewhere between novice and serious quant: Mean reversion on highly correlated pairs.

The strategy you can stick to for 12 months beats the strategy you'll abandon after a 20% drawdown.

More reading

• Bot Examples Library — runnable starter code
• Kalshi Market Making — deep dive
• How to Backtest a Trading Strategy
• Why Your Trading Bot Keeps Crashing