LEAPS are options with roughly a year or more to expiration, used to express multi-month directional theses with a fraction of the capital of stock ownership. The standard convention runs deep in the money, around 0.70-0.75 delta, where the option behaves mostly like stock and time decay is gentlest per day.
The lottery-ticket structure fights its holder twice: an out-of-the- money LEAPS is mostly time value, so it pays maximum decay, and its lower delta means the thesis must be very right before the position moves. At 0.70-0.75 delta the option is mostly intrinsic value; it tracks the stock at three-quarters speed for perhaps a third of the capital, decay per day is small, and a merely-correct thesis pays. The implicit leverage still cuts both ways, which is why worst-case sizing treats the full premium as the loss to budget.
Two that short-dated traders ignore. Vega: a two-year option carries large sensitivity to implied volatility, so entering when volatility is cheap matters as much as entering when the chart agrees; a thesis that is right about direction can still tread water against a volatility decline. Rho: rate sensitivity compounds over years, which is why careful pricing engines discount each expiration off a real yield curve rather than one constant rate. Early-exercise mechanics also matter more, since dividends interact with deep in-the-money calls; pricing that strips the early-exercise premium keeps the quoted implied volatilities honest.
A documented anomaly with a clean options expression: stocks that report large positive earnings surprises with strong same-day moves tend to keep drifting in that direction for weeks. A screen for it requires a meaningful surprise (for example EPS beat above 10%), a strong announcement-day move, a price still near the announcement level, and, because the position buys options, volatility that is not already inflated. The expression is the LEAPS convention above at a 9-to-12-month expiration: long enough for the drift to play out, in-the-money enough that the position needs drift rather than a moonshot. It is a screen, not a promise; the anomaly's edge is statistical and the sizing rules treat every instance as capable of its full loss.