Risk Management
Your Option Spread Costs 8.6x the Stock Tick — Every Timeframe
The options bid-ask spread is the largest number in short-term options trading and almost nobody measures it against the thing it has to be paid out of. Quoted in cents it looks trivial: three cents on a sixty-cent contract. Converted into R — multiples of the risk unit an edge is measured in — that same three cents costs between 0.06 and 0.27 R per round trip. The largest edge produced by any hypothesis in a 48,836-signal research program was +0.03 R, and it was not statistically significant.
That is the whole picture in two numbers. The cost is an order of magnitude larger than the edge, and unlike the edge it is entirely under your control.
Why cost in R is the only comparison that means anything
Cost in dollars cannot be compared to expectancy. Cost as a percentage of premium cannot either — it ignores how much underlying movement the contract actually captures. Both must be pushed into the same unit as the edge.
The conversion is two divisions:
Dividing by delta turns option dollars into underlying dollars: a $0.03 spread on a 0.35-delta contract is a $0.086 move in the stock you must win back before you are flat. Expressing that as basis points of price and dividing by 1R — the ATR of the timeframe you trade — gives a number directly comparable to any expectancy figure.
The option spread is 8.6x the underlying tick — at every timeframe
The tick is the cost most traders internalise, because equity trading taught it to them. It is not the binding cost when you trade options.
| 1m | 5m | 15m | |
|---|---|---|---|
| 1R = ATR(14) on that timeframe | 8.7 bps | 23.1 bps | 40.3 bps |
| one-tick round trip, in R | 0.032 | 0.013 | 0.007 |
| $0.03 option spread, in R (δ=0.35) | 0.273 | 0.108 | 0.061 |
| + decay drag at 5%/hour of premium | 0.330 | 0.270 | 0.406 |
| gross expectancy measured, both sides | −0.028 | −0.053 | −0.047 |
Both the spread and the tick scale as one over the square root of the timeframe, so the ratio between them never changes. Raising your timeframe does not improve the option relative to the underlying. It is 8.6× at one minute and 8.6× at fifteen.
The cost floor has an interior minimum near 5 minutes
The obvious response — trade a slower timeframe, 1R grows, the cost fraction shrinks — is half right, and the half that is wrong is the expensive half.
The spread term does fall as timeframe rises, at tf^−½. But the decay term rises at tf^+½, because a slower signal means a longer hold and longer holds pay more theta and more variance risk premium. Add them and the total does not go to zero. It bottoms out around the five-minute mark at roughly 0.17–0.27 R and rises again after that.
Three hypotheses tested, none survived
The cost floor only becomes decisive once you know how large the edges being defended actually are. Three were tested at scale, pre-declared, and reported whether they worked or not.
| The VWAP gate | Higher timeframe | Trend alignment | |
|---|---|---|---|
| Claim | only short below session VWAP | bigger 1R clears the cost floor | don't fight the day's direction |
| Result | refuted | refuted | not supported |
| Sample | n = 19,452 | n = 48,836 | n = 38,897 |
| Effect | +0.003 R, p = 0.85 | cost falls 4.4x, expectancy does not | +0.027 R, Holm p = 0.34 |
| Twist | inverts within-session to −0.391 R | negative at 1m, 5m and 15m | smaller than the tick it costs |
The VWAP gate is the instructive failure. It looked overwhelming in a single session — one symbol printed 0 of 8 signals below VWAP, another 9 of 10 — until the obvious question was asked: on 33.6% of sessions every signal lands on one side of VWAP anyway. The stunning contrast was the modal session, not a discovery. Tested within sessions rather than across them, it collapses, and the sign flips.
Why does a within-session test change the answer so completely?
Because a variable can be 51% a day-level property. "Below VWAP" mostly encodes what kind of day it is, which you learn only after the day is over — it is not information available at the signal bar. Split the effect into its between-session and within-session parts and they point in opposite directions: +1.086 within, −0.519 between, with the between component carrying 69% of the variance. They cancel. Any gate you propose has to be tested inside sessions, or you are measuring hindsight.
The structure signal underneath all three is negative before a cent of cost at every timeframe tested, with confidence intervals excluding zero on the wrong side. There is nothing there for a filter to rescue, which is precisely why knowing what a real edge is matters more than accumulating conditions.
What actually reduces the cost
Since no filter has ever been worth more than 0.03 R and the spread is worth 0.06–0.27 R, attention belongs on the big number. Four levers, in order of leverage:
- Work a mid-limit, always. The single highest-value habit here. If the order will not fill at mid, the trade was not good enough to justify paying the spread — that is a filter, and unlike the tested ones it is free.
- Raise delta. Cost in R divides by delta. Moving from 0.15 to 0.45 delta cuts the same spread's cost in R by two thirds, and buys reachability at the same time.
- Hold shorter, but enter early. Decay is time × premium. The first 60–90 minutes give the most realised volatility per unit of spread paid.
- Trade less often. Every round trip is a guaranteed cost against an expectancy that is not positive. Frequency multiplies the one number you know is real.
The takeaway
Stop hunting for the condition that turns the strategy positive. Three well-powered attempts found nothing above +0.03 R, and the cost of expressing any of them is two to nine times larger than the effect itself.
Price your round trip in R before you take it. If the cost exceeds the largest edge you have ever measured, the fill is your strategy — and improving the fill is the only work in this whole problem with a guaranteed payoff.