"Algorithmic trading" sounds like a single skill. It isn't. Underneath the label sit a handful of distinct families, and what separates them isn't the code — it's the answer to one question: why does this return exist at all? Each family below gives a different answer, runs as a real tool on the platform, and carries numbers from its own backtest. And each has a specific environment where it stops working — stated up front, because a strategy you can't break on paper is one you don't yet understand.
1. Trend following & momentum
Its bet is that information is priced in gradually and that money chases performance, so a move that has started tends to run longer than it "should." The rule is mechanical: a 50/200-day EMA crossover marks a confirmed trend, the position is sized to EWMA volatility at a 25% annual target, and a 2.5×ATR(20) trailing stop takes it out. On a decade of Bitcoin it earned a Sharpe of 1.47 across 195 trades — with only 44% of them winners, because the winners are simply far bigger than the losers. Breaks when markets turn sideways: crossovers fire in both directions and the false signals pile up. (Adaptive Trend.)
2. Mean reversion
The opposite bet: price drifts away from a reference value on enthusiasm or panic, and a kind of statistical gravity eventually pulls it back. Here price relative to its 20-week moving average is mapped onto eight risk zones — as risk falls, buying scales up to as much as ×20; as euphoria builds, the position is trimmed. It produced a Calmar of 2.49 from 11 buys, all of them clustered at the 2022 bottom. Breaks when the regime itself shifts and what looked "cheap" quietly becomes the new normal. (Short-Term Bubble Risk.)
3. Relative value
Instead of a price, it bets on a relationship: two closely related assets tend to return to their historical ratio, even when both are rising or falling together. A rolling channel (mean ± 2σ) is drawn on the gold-to-silver ratio over 260 weeks; at each edge the entire holding switches into the cheaper metal, so the ounce count compounds without adding a dollar of new money — three switches in ten years grew the inventory by 96.1%. Breaks when the relationship itself breaks, on a real industrial or monetary shift. (Gold/Silver Ratio.)
4. Contrarian sentiment
Its premise is behavioral: the crowd reliably buys tops and sells bottoms, so quantifying emotion puts you on the other side of that trade. It buys 5% of capital each time the Fear & Greed index drops to ≤ 25 and trims 20% when it climbs to ≥ 75, with RSI and EMA-200-deviation filters so it doesn't start buying a fall too early. Over 736 "fear" days it bought $7,343 and sold $23,217. Breaks when the fear turns out to be justified and simply lasts — a genuine bear market that keeps paying you to be wrong. (Fear & Greed DCA.)
5. Accumulation design
This family doesn't bet on an asset at all — it bets that how you enter matters as much as what you buy. The same $10,000 is run four ways in parallel on one asset over one period: a single lump sum, equal weekly instalments, dynamic risk-weighted instalments, and a benchmark. On a trending asset the paths diverge hard — $118,576 versus $88,935 versus $81,689. Breaks when the asset has no uptrend at all: with nothing to compound, clever entry design just spreads the same flat outcome around. (DCA Strategy Lab.)
6. Portfolio construction
The last family bets on combination: return comes from how assets sit together, and low correlation cuts risk faster than it cuts return. It generates tens of thousands of possible allocations, draws the efficient frontier, and extracts the maximum-Sharpe portfolio — then decomposes each holding's contribution to risk, not to weight, because a 20% position can quietly carry half the danger. The max-Sharpe mix reached a Sharpe of 1.44 against 0.79 for the naive alternative, from just 1.8 truly independent bets. Breaks when a crisis drives every correlation toward one and the diversification you were paid for evaporates. (Cross-Asset Risk.)
The families at a glance
| Family | Why the return exists | Real result | Breaks when |
|---|---|---|---|
| Trend following | Trends persist longer than logic suggests | Sharpe 1.47 · 195 trades | Sideways markets, false crossovers |
| Mean reversion | Price overshoots on emotion, then reverts | Calmar 2.49 · 11 buys at the 2022 low | Regime shift — "cheap" becomes normal |
| Relative value | A pair reverts to its historical ratio | +96.1% inventory · 3 switches / 10 yrs | The relationship itself breaks |
| Contrarian sentiment | The crowd is wrong at extremes | Bought $7,343 / sold $23,217 | Fear is justified and lasts |
| Accumulation design | How you enter matters as much as what | $118,576 vs $88,935 vs $81,689 | The asset has no uptrend |
| Portfolio construction | Return from the combination, not the asset | Sharpe 1.44 vs 0.79 · 1.8 bets | Correlations jump to 1 in a crisis |
No family wins forever
Read the six "breaks when" lines together and the real lesson appears: none of these is the strategy. Each earns in one environment and bleeds in another — trend following pays in strong directional markets and gives it back in the chop; mean reversion is brilliant into a washout and disastrous into a regime change. So the quant's job was never to find the single best system. It is two quieter disciplines: knowing which environment you are actually in right now, and knowing in advance exactly what would prove the hypothesis wrong if that environment changed. Every "breaks when" above is that tripwire, written down before the money is on the line.
That is also why the rest of this series turns to measurement rather than more strategies: how to tell whether an edge is real, which return figure you should even look at, what each risk-adjusted metric quietly hides, and the ways a backtest can lie to you.