Peak-to-Close Give-Back: The One Number Your Options Backtesting Software Does Not Show You

Open almost any backtest report for an expiry-day option-selling strategy and you get a table of days. Date, entry time, exit time, net points, net rupees. Maybe a cumulative equity curve stitched from those daily closes. Add it up, divide by the number of days, and you have an average day profit you can quote.

That table is a summary of where each day ended. It says nothing about where each day went. And for a short options position, where the day went is most of the story, because the trade is marked to market every minute you hold it and your own nerve is marked to market along with it.
Here is the measurement that makes this concrete. In a study of 163 NIFTY and SENSEX weekly expiry sessions between January 2025 and July 2026, built on 1-minute option-chain data, a plain short at-the-money straddle entered at 09:20 was tracked minute by minute to the close. On the 159 sessions with usable intraday paths, the median day gave back 42.8% of its peak intraday profit before it closed. Not 4%. Not 14%. Over two-fifths of the best number the position ever printed, handed back by 15:20.
That number does not appear in a net-P&L table anywhere, by construction. This article is about what it means, why it is the first thing you should ask any backtesting tool to show you, and what a report has to expose before you can honestly say you have tested an option-selling strategy rather than just scored it.

What peak-to-close give-back actually measures

Take one trading day. You sell a straddle at 09:20. From that moment your position has a mark-to-market value every minute until you exit. Plot it. You get a line that starts at zero, wanders, and ends at whatever the day’s net P&L turned out to be.
Two points on that line matter beyond the endpoint. The highest point the line ever reached — the peak unrealised profit — and the gap between that peak and the close. Express the gap as a share of the peak and you have the give-back ratio for the day:

Give-back % = (peak MTM profit − closing P&L) ÷ peak MTM profit

A day that peaks at 40 points of profit and closes at 20 gave back 50%. A day that peaks at 40 and closes at 38 gave back 5%. Both are perfectly ordinary outcomes. In a net-P&L table they look like a bad day and a good day. In reality they are the same strategy behaving in two completely different ways, and the second one is far easier for a human to hold.
Note what give-back is not. It is not maximum adverse excursion, which measures how far underwater you went before recovering. Those two capture different fears. Adverse excursion is the fear of the stop. Give-back is the subtler and more corrosive one: watching money you already counted leak away while you sit on your hands because the rules say hold till 15:20.

The desk numbers, and what the distribution looks like

Here is the full set of give-back figures from those 159 usable expiry sessions. One trade per day, short ATM straddle, 09:20 entry, held to the session’s scheduled exit, minute-level marks throughout.
MetricValueMedian give-back, as % of peak profit42.8%Mean give-back, % of peak (winsorized at 100%)50.0%Days giving back more than 30% of peak86 of 159 (54%)Days giving back more than 50% of peak78 of 159 (49%)Days where a positive peak closed as an outright loss48 of 159 (30%)
Read those five rows together and a picture forms that a daily P&L column cannot give you.
More than half of all sessions — 54% — surrendered at least 30% of their best number. Almost half, 49%, surrendered more than half. And the row that should stop you: on 30% of days, roughly three sessions in ten, the position was in profit at its peak and still closed in the red. Not “closed with a smaller profit”. Closed as a loss, after having been a winner at some point in the day.
Think about what that does to a trader’s experience of the same strategy whose annual numbers look acceptable. Nearly one day in three, they watch a live gain turn into a realised loss. The equity curve may still slope up. The lived experience is of a strategy that repeatedly takes back what it gave, and that experience is what drives the decisions that actually destroy the account.

The gap between the strategy and the trader

Backtests assume an obedient operator. The rule says exit 15:20, so the backtest exits 15:20, every single day, without flinching, for 159 consecutive sessions. Nobody does this. What people do instead, when they are 42% off the peak at 14:30 and still bleeding, is one of three things:

• They book early. They take the profit at the first sign of fade. This feels like discipline. It converts the strategy into a different strategy with a different and untested distribution, usually one that clips winners and keeps full-sized losers.
• They over-manage. They adjust, hedge, roll a strike, add a leg. Each intervention is an untested decision made under stress. The backtest they are relying on now describes nothing they are doing.
• They quit the system. After the fourth or fifth session in a row where a good peak became a bad close, they conclude the edge is gone and stop. Very often they stop right before the run that would have paid for the drawdown.

All three are behavioural failures. All three are predictable from the path data and invisible in the summary table. This is why the path is not a nice-to-have diagnostic — it is the input to the question of whether a human can actually run the thing. A strategy with a 10% median give-back and one with a 43% median give-back can post the identical average day. They are not equally tradable, and no amount of net-P&L reporting will tell you which one you are holding.

Why identical P&L does not mean identical risk

Let us make the arithmetic explicit, because this is where most people nod along and then go back to ranking strategies by average profit.
Two configurations, A and B. Both are tested over 100 expiry days. Both net an average of 20 index points per day on one lot. NIFTY’s lot size is 65, so 20 points is Rs 1,300 per lot per day, and across 100 days both show about Rs 1,30,000 on a single lot before costs. Same mean, same total, same line on the equity chart if you only plot daily closes.
Now the paths. Strategy A’s typical day peaks at 23 points and closes at 20 — a 13% give-back. Strategy B’s typical day peaks at 44 points and closes at 20 — a 55% give-back. B’s trader sees 44 points of profit on screen, repeatedly, and ends the day with 20.
Ask yourself which operator survives a bad month. A’s trader has almost nothing to second-guess; the close is close to the best available outcome, so holding to the rule feels obviously right. B’s trader has 24 points of visible, countable regret every single session. B’s trader will start exiting at the peak. The moment they do, they are no longer running strategy B, and the 100-day backtest on which they sized their capital no longer applies to them in any form.
There is a second, harder consequence. A deep give-back profile usually means the position is carrying real directional exposure late in the session — the strike was fixed in the morning and spot has walked away from it, so one leg has gone from a theta earner to a gamma liability. The give-back is the symptom. The exposure is the cause. A strategy that gives back half its peak on half its days is telling you, in a language the summary table cannot speak, that its risk profile changes shape during the day. Tools built for the Indian weekly-expiry market, like the options backtesting software built for the Indian options market used on index-option-selling desks, earn their keep precisely by storing the minute-level marks that let you see this, rather than collapsing each session to a single number.

The premium curve explains the shape of the path

Give-back is not random noise. It has a structure, and the structure comes from how expiry-day premium actually behaves. The same 163-session dataset gives us the average ATM straddle premium through the day, indexed to the 09:16 premium as 100.
Time% of entry premiumTime% of entry
This is the opposite of the curve most expiry-day explainers draw. The popular picture is a smooth slide toward zero, steepening into the afternoon. What the data shows is a fast drop in the first two hours — 100 down to 88.6 by 11:16 — and then a sideways chop in an 88 to 93 band for the rest of the session. The average does not grind lower after lunch. It mills about and even ticks up.
The mechanism is straightforward once you say it out loud. The strike is fixed at the 09:20 ATM level. Spot does not stay there. As the session goes on, the series you sold drifts away from the money, and what you collect in time decay is netted against what you lose when spot walks back toward or through your strike and the option re-prices. By the afternoon you are not holding a clean theta position. You are holding theta minus gamma re-pricing, and on a decent fraction of days gamma wins the afternoon.
That single fact — front-loaded decay, then a flat and noisy afternoon — is where the 42.8% median give-back comes from. Most of the easy money is earned before lunch. Everything after that is a coin-weighted bet that spot does not come back for your strike. The peak, unsurprisingly, often lands in that late-morning window, and the close is a draw from the afternoon chop.

Why your exit rule changes the path more than it changes the average

This is the practical payoff. Consider two exits from the same 09:20 entry: one at 13:00, one at 15:15. On the average curve, premium is 89.1% of entry at 13:00 and 91.3% at 15:20. So the early exit captures a slightly larger share of the decay than the late one, on average — the later exit is holding through a band that, in the mean, does not fall any further.
But the averages understate what actually changed. The early exit cuts the position out of the entire window in which gamma re-pricing produces deep give-backs. The late exit sits in it for two and a half hours. Two rules on the same entry, nearly the same mean, radically different path profiles — and so radically different behaviour under live conditions, different intraday drawdowns, and different odds of a profitable peak closing red.
Any report that compares those two rules on average profit alone will tell you they are nearly the same strategy. They are not. If you are weighing one expiry-day tool against another, how each one lets you compare exit-time variants on path, not just on mean, is worth more than the length of its strategy library. This is the sort of thing a serious comparison of expiry-day backtesting tools should be judged on, and it is routinely left out because it is harder to chart than a bar of average profit.

Median, not mean — and why that is not a technicality

Notice the brief gave two numbers: a median give-back of 42.8% and a winsorized mean of 50.0%. The median is the reportable figure. Here is why, and the reason is specific to this metric rather than general statistical hygiene.
The give-back ratio has peak profit in the denominator. On a day where the position barely got into profit at all — a peak of half a point, say — the ratio explodes. Peak 0.5 points, close −6 points, and your give-back comes out at 1,300%. The day was unremarkable. The ratio is absurd. Average a handful of those into 159 observations and the mean is no longer describing the strategy; it is describing the handful of near-zero denominators.
Two defences exist. The median ignores the magnitude of outliers entirely and reports the middle day, which is exactly the day you want described. Or you winsorize — cap each observation at 100%, so “gave back everything and then some” counts as 100% and nothing more — and then take the mean. That cap is what produces the 50.0% figure. Both are legitimate. Quoting a raw, uncapped mean is not, and if a report gives you a single average give-back number without telling you how it handled the denominator, treat the number as unverified.
The deeper lesson is about distributions. “Median 42.8%” is one fact. The four threshold counts beneath it — 54% of days over 30%, 49% over 50%, 30% flipping to a loss — are what make it usable, because they tell you how often you will be in the uncomfortable region rather than what the typical discomfort was. A report that shows you only a central tendency for give-back has given you the least informative summary of a quantity whose whole importance lies in its tail. Ask for the distribution. A histogram or a set of deciles is not a large thing to want.

The path metrics an option-selling backtest report must expose

If you are evaluating what a tool reports, or specifying what you want built, this is the minimum list for short-option strategies. None of it is exotic. All of it falls out of a stored minute-level MTM series.

MetricWhat it answersWhy the summary table cannotMTM time series, per dayWhere did the position actually go?It is the raw material for everything below; without it nothing else is computablePeak unrealised profit, with timestampHow good did it get, and when?Net P&L records only the final markMax adverse excursion (MAE)How far underwater before recovery?A day that closed +15 may have been −60 at noonPeak-to-close give-back, full distributionHow much of the best number survives to the close, and how often is it severe?Needs both peak and close; a mean alone hides the tailIntraday drawdown from peak, in rupeesWhat is the real-money size of the fade, not just the ratio?A 60% give-back on a 3-point peak is noise; on a 70-point peak it is a crisisCount of peak-profit-to-loss flipsHow often does a winner close as a loser?Pure path event; no trace of it survives in the daily columnTime-in-profit, as % of holding periodWas the day comfortable or a knife fight?Two days closing at +20 can differ completely herePeak-time histogramDoes the strategy top out at 11:00 or hold into 15:00?Tells you whether your exit rule is leaving money or collecting risk
Two of those deserve a note. Intraday drawdown from peak in rupees matters because the percentage ratio alone is scale-blind, and scale is what triggers human intervention. Nobody panics over a 70% give-back of Rs 800. Everybody panics over a 35% give-back of Rs 45,000. Report both the ratio and the absolute.
The peak-time histogram is the one most often missing and the one that most directly informs exit design. If peaks cluster before noon, a 15:20 exit is systematically donating the afternoon chop back to the market. If they cluster after 14:30, an early exit is leaving the best part of the day on the table. You cannot know which without the timestamps, and you cannot get the timestamps from a summary row.

Getting path data out of a tool that only gives you a summary

Most retail platforms will not hand you a minute-level MTM series. There are still routes to a usable approximation, in rough order of fidelity.

1. Check for a trade-log or tick export first. Many tools expose a per-leg log with entry and exit prices and timestamps but bury it behind a download button nobody clicks. If you can get leg-level fills plus a time series of option prices, you can reconstruct MTM yourself.
2. Use the exit-time sweep as a proxy. Run the same strategy with the exit time stepped every 15 minutes from entry to close. The resulting series of average P&L by exit time is a crude mean MTM curve. It gives you the shape of the average day, which is enough to spot front-loaded decay or an afternoon bleed. It will not give you per-day peaks or the flip count, because averaging across days smears away the individual paths.
3. Reconstruct from option-chain data. If you hold or can obtain 1-minute option prices for the strikes you traded, MTM is arithmetic: for a short leg, mark-to-market equals entry price minus current price, times lot size, times lots, summed across legs. This is the only route that gives you true per-day paths, and it is why desks that care about this keep their own chain data rather than relying on a vendor’s summary.
4. Instrument your own forward test. If a tool runs live or in paper mode, log the position value yourself on a one-minute timer. A few weeks of this will not validate a strategy, but it will tell you very quickly whether the fade you are worried about is real.

If none of these are available — no log export, no exit-time sweep, no chain data, no paper-mode logging — then you cannot measure the path at all with that tool, and you should be honest with yourself about what your backtest has and has not established.

How to judge an options backtesting tool on path quality

Set aside the feature lists. For option selling specifically, three questions sort the serious tools from the rest, and all three can be answered in a trial period.

1. Can you export the MTM path at all?

Not the equity curve of daily closes. The within-day mark-to-market series for a single session, as data you can open in a spreadsheet. If the answer is no, every metric in the table above is out of reach and you are permanently restricted to scoring strategies on their endpoints. Ask for the export before you subscribe, not after.

2. At what granularity?

Granularity is not a detail; it determines whether the peak you measure is the real peak. A 1-minute series on an expiry-day straddle will find a peak that a 15-minute series misses entirely, because the move that produced it took six minutes. End-of-day-only marking is useless here. Hourly is close to useless. 5-minute is workable for broad comparisons. 1-minute is what you need if give-back and MAE are going to be decision inputs. And ask what the mark is based on — last traded price on an illiquid strike can print a peak that no one could have transacted at, so a tool that marks off the bid-ask midpoint is telling you something more honest than one that marks off a stale LTP.

3. Does it model the option chain, or proxy the index?

This is the question that separates tools that can answer path questions from tools that merely appear to. Some engines do not hold historical option prices at all. They simulate the index, then price the options from a model using an assumed or interpolated volatility. For a directional index strategy this is defensible. For a short straddle whose entire path is driven by how real premium re-priced against a fixed strike, it is not. The India VIX relationship in the same dataset shows why: ATM straddle premium as a share of spot runs 0.38% in the calmest VIX band and 0.83% when VIX is above 20 — more than double. A tool that proxies volatility instead of reading the actual chain will get the level wrong, and will get the intraday re-pricing wrong in a way that systematically understates give-back, because the messy part is exactly what a smooth model smooths out.
Ask directly: are these historical traded option prices, or model-derived? If the answer involves an implied-volatility assumption, your give-back numbers are a property of the model, not of the market. The exchanges publish historical derivatives data and settlement information through their own channels — NSE for NIFTY and the BSE for SENSEX — so there is no good reason for an India-focused tool to be guessing at option prices that were publicly recorded. If you want the textbook treatment of how options are marked and what mark-to-market means mechanically, Investopedia’s explainer on mark-to-market is a reasonable starting point.

A fourth question, if you want to be thorough

Does the report let you compare path metrics across variants side by side? Measuring give-back on one configuration is diagnostic. Being able to sort twenty configurations by median give-back, with average profit held roughly constant, is a design tool. That is the point at which path data stops being a curiosity and starts changing which strategy you deploy — you can deliberately select the variant that a human being can actually sit through.

What this does not tell you

Three honest limits on everything above.
First, the figures describe a specific instrument on a specific market in a specific window: NIFTY and SENSEX weekly expiry, 09:20 ATM straddle entry, 163 sessions across January 2025 to July 2026, of which 159 had usable paths. Other entry times, other structures, other periods will produce different give-back distributions. The method transfers. The numbers do not.
Second, a low give-back is not automatically a better strategy. A position that never builds much unrealised profit has little to give back, which is arithmetically flattering and tells you nothing good. Give-back is a ratio and ratios need context. Always read it next to the absolute rupee peak and the net result.
Third, this is in-sample measurement on historical data. Minute-level reconstruction says what the marks were, not what you could have transacted at. Slippage, spread on a far strike in the last half hour, and the gap between a model fill and a real one all sit on top. Path analysis makes a backtest more honest. It does not make it live trading.

A note on risk and what this article is

Options selling carries a real and open-ended risk of loss. The give-back data above is itself a demonstration of that: on 30% of the sessions studied, a position sitting in profit at its peak closed as a loss. Backtested results, however carefully reconstructed, are historical measurements and are not a forecast of future outcomes. Any tool mentioned here, EliteAlgo included, is analytical software and not an investment adviser; nothing in this article is investment advice or a recommendation to trade any strategy or instrument. Size positions on capital you can afford to lose, and treat every number here as something to verify on your own data rather than something to act on.

Frequently Asked Questions

What is peak-to-close give-back in simple terms?
It is the share of your best unrealised profit for the day that you hand back before the position closes. If the trade was up 40 points at its best and finished up 20, the give-back is 50%. It is calculated per day from the minute-by-minute mark-to-market value, so it needs path data, not just the closing P&L.
Why does my options backtesting software not show the MTM path?
Usually because storing and serving a minute-level series for every day of every backtest is expensive, and because most users never ask for it. Some engines also do not hold historical option prices at all — they simulate the index and price options from a volatility model, in which case there is no real path to show you. Ask whether the prices are historical traded prices or model-derived before you rely on anything path-based.
Does a high give-back mean the strategy is bad?
No. It means the strategy is hard to sit through, which is a different and sometimes more important problem. A strategy can post a healthy average day and still be unrunnable by a human because they will intervene during the fade. Judge give-back on tradability, then check whether an earlier exit or a different structure keeps most of the profit with less of the path pain.
Is 1-minute data really necessary, or will 5-minute do?
Five-minute is usable for broad comparisons between variants. One-minute is what you want if give-back or max adverse excursion will drive a sizing or deployment decision, because coarse sampling misses peaks and troughs that lasted only a few minutes — and on expiry day those are exactly the moves that matter. End-of-day marking tells you nothing about the path at all.
Which exit time should I use if premium stops decaying after 11:00?
The data argues for testing early-afternoon exits properly rather than defaulting to the close, because on average the premium level barely moves between roughly 11:15 and 15:20 while the position keeps carrying gamma risk. But “on average” is doing a lot of work in that sentence, and the right answer depends on your structure, your strike selection and your costs. Run the exit time as a parameter, compare the variants on path metrics as well as mean profit, and paper trade before committing capital.
Can I compute give-back myself if my tool only gives me a trade log?
Yes, if the log has leg-level entry prices and timestamps and you can source 1-minute prices for those strikes. For a short leg, mark-to-market at any minute is entry price minus the current price, multiplied by lot size and number of lots, summed over legs. Take the maximum of that series for the peak and the final value for the close, and you have the day’s give-back. The work is in sourcing clean option prices, not in the arithmetic.

Conclusion

A net-P&L table ranks strategies by where they ended. It is the right tool for that job and the wrong tool for almost every other question an option seller needs answered. The path — where the position went, how good it got, when it got there, and how much of that survived to the close — is where the risk actually lives, and it is the part that decides whether a strategy with acceptable numbers is one a person can run for a year without abandoning it.
The headline figure is worth repeating because it is so far from what most reports imply. Across 159 usable NIFTY and SENSEX expiry sessions, the median day gave back 42.8% of its peak profit, more than half of all days gave back over 30%, and three days in ten turned a profitable peak into a closing loss. None of that is visible in a daily P&L column, and none of it is unknowable. It is a stored minute-level series and some arithmetic.
So when you next sit down with a backtesting tool, before you look at the average day profit, ask the three questions. Can I export the mark-to-market path? At what granularity? And is it reading the real option chain or a model’s idea of one? If the answers are no, coarse, and a model, then the pretty equity curve on the screen is a summary of endpoints and nothing more — and you still do not know how the days you are about to live through will feel.