Introducing backfin structured notes
The payoff, and what the issuer kept.
A market-linked debenture is a bond and a handful of options in a wrapper. We take the wrapper off, price each leg from first principles, discount the bond at the issuer’s own credit spread, and put the result beside the issue price. The distance between them is the margin.
As sold · capital-protected NIFTY note · AA− issuer
100% protected.
1.35× the upside.
“Capital protected, with leveraged upside on the index.”
3real 2026 term sheets, torn down and re-priced
2%–30%retained margin across them, same words to the client
0.01per 100 · the gap between two independent methods
712automated tests behind the engine
Fair value is a band. The issue price is a point. The distance between them is what the client paid for the wrapper.
Three notes · by rating only
Same shelf. Same words to the client.
Three market-linked notes offered on the street this year, all sold as protected or enhanced exposure to an index. Priced the same way, discounted at each issuer’s own rating-implied spread. Issuers are identified by rating and never by name.
AA− · NIFTY · 3.5 years · terminal
93–98fair value per 100 · issue price 100
A zero-coupon bond and 1.35 at-the-money calls. Nothing hidden in the structure; the margin is the cost of the wrapper.
AA− · NIFTY · 2.08 years · cliquet, two annual resets
96–99fair value per 100 · issue price 100
7% floor, one-for-one from +7% to +16%, knocked back to 7% above that. The entire equity kicker is worth 1.5 points. In 57% of years it pays the floor; expected return about 8% a year against a marketed maximum of 15.26%.
A · SENSEX · 5.31 years · average of 7 observations
70–74fair value per 100 · issue price 100
Not principal protected: a digital, a call spread and geared puts below −15%, on an A-rated bond. Averaging does the quiet work: the index finishing above +34% is a 71% event, the seven-observation average clearing it is 50%.
Two hundred thousand paths
What the client actually gets.
A brochure prints a maximum. The engine prints the distribution: the same Monte Carlo that prices the note, read as the return the client takes home at maturity, with the tails marked.
AA− · 3.5 years · capital protected · 1.35×
0%probability of a loss
+18%median return at maturity
7.6%expected, a year
The floor does what it says: no path loses. The cost of the floor is the left half of the distribution piling up at zero.
A · 5.31 years · averaged · not protected
13%probability of a loss
+0%median return at maturity
4.4%expected, a year · risk-free is 6.5%
Averaging seven observations narrows the upside the brochure draws and leaves the geared downside where it was. The expected return sits below the risk-free rate before the issuer’s margin is counted.
Underneath the price
Value, delta, and the surface it came from.
A note’s value moves with the index before it ever pays out, and every option leg in it is priced off an implied-volatility surface. Both are shown as the engine computes them: the value curve from the pricer, the surface fitted by the desk to a recorded NIFTY option chain.
fair value per 100 · against the index at inception
NIFTY implied volatility · by expiry · recorded chain, 10 Sep 2026, 11:00
The smile is refitted every twelve minutes from the recorded chain and plays through the session, 09:26 to 15:29; drag to stop it on a time. The five-day slice is the steep one: that is what a weekly NIFTY smile looks like. Dots are the raw inverted quotes at eleven; lines are the fit. The chain is recorded every session by the desk’s own recorder, because a chain that is not stored today is gone.
Two methods, or no number
Priced twice. Published once.
Every note is priced two independent ways. A direct Monte Carlo of the full payoff, and a decomposition of the piecewise payoff into vanilla legs, each priced in closed form and summed. They have to agree, or the note is not priced.
risk-free value per 100 · before the issuer’s credit
| note | decomposition | Monte Carlo | difference |
|---|---|---|---|
| AA− · 3.5y · terminal | 102.87 | 102.88 | 0.01 |
| AA− · 2.08y · cliquet | 101.49 | 101.48 | 0.01 |
| A · 5.31y · averaged | 88.91 | 88.91 | 0.00 |
Decomposition exactness is checked to one part in a hundred million before either price is trusted. Sampling error below 0.02 per 100 on the reconciliation, below 0.1 on the credit grids. Up to two million paths.
the A-rated note, taken apart · per 100
Risk-neutral return about 4.3% a year against a 6.5% risk-free rate. Then the issuer’s own spread takes it to 70–74.
It also runs the other way
What you were offered. What was fair.
Fix any two of participation, coupon and barrier, and the third is solved to par, or to a margin you choose. A term sheet becomes a negotiation rather than something you accept or decline.
solve_third · to par · the three notes
The number to take back to the issuer. Or the number that ends the conversation.
the flat-market test · AA− 3.5y note · what the client actually gets
0%index flat at maturity
+13.5%index +10%
+29%expected · 7.6% a year
A deterministic grid at any index move, and the full Monte Carlo distribution behind it: median, the 5th to 95th percentiles, the probability of a loss and of finishing at par. The question a client asks first, answered before the pitch.
The desk twin · live NIFTY data · 2026-08-25
The other side of the trade.
The same engine runs an issuer’s book: fifteen notes on the live NIFTY feed, hedged, stressed and attributed every day. It is how we know what a note costs to run, not just what it is worth.
hedging error · standard deviation of P&L · by rebalances a year
From 285 at five rebalances to 90 at daily. The dashed line is what theory says, one over root n; the gap above it is gamma the model does not see.
stress · book P&L as a percent of NAV
Equity crash dominates; an issuer-spread widening of 50 basis points costs a third of a percent. The order matters more than the numbers.
rolling issuance · one-year autocall · NIFTY history
109issuances, rolled through the history
Model reserve across the book: stochastic-local volatility values the fifteen notes 12 basis points away from local volatility, and that difference is held, not ignored. Hedge notional is 36 crore against 12,000 crore of daily volume: 21 minutes to hedge at 20% participation.
By construction
A band, never a point.
Guaranteeda word the engine cannot emit · nor assured, nor fixed return
93–98every fair value is a band across the volatility range, discounted at the issuer’s own spread
verdicts: clears · borderline · fails · against a margin threshold the desk sets
The term-sheet valuer prices only what it can decompose and check two ways: terminal, averaged, barrier and worst-of. Autocallables, knock-ins and lookbacks come back as a refusal, not a number. The desk twin prices them by Monte Carlo under local and stochastic-local volatility, and holds the difference as a reserve.
The principal leg is discounted at the rating-implied spread, AA− at 200–275 basis points, A at 350–450, so a weaker name is never flattered by a government curve.
Worst-of prices are not yet reconciled with issuers’ own disclosed values. Local volatility, dividends and funding have each been tested; none closes the gap, and the row stays open on the page until one does.
Checked against something outside
The model, the issuer and the option market agree.
US issuers must publish their own initial estimated value in an SEC pricing supplement. Inverting one for volatility, at a real bank funding level, is the closest thing to an answer key that exists.
SEC 424B2 · a US bank’s single-stock note
48.7%volatility implied by the issuer’s own estimated value
48.1%the listed one-year at-the-money quote on the same stock
Inside one volatility point, from three independent sources.
the US shelf · 600 recent supplements
67%are worst-of baskets on single stocks
So correlation is the binding unknown, and the engine prices worst-of across a correlated basket with the matrix repaired to be valid before it is used. Implied equicorrelation read from the street: 0.74–0.95, an upper bound rather than a measurement.
an issuer’s book · five notes
81 → 49gross legs netted to what actually needs hedging
Decompose each note, net the legs across the book. Forty percent cancels internally. The case for warehousing rather than hedging note by note, in one number.
Where it sits
Not the issuer’s sheet. Not a spreadsheet.
The two ways a desk decides on a note today: the issuer’s own pricing sheet, taken on faith, or a spreadsheet that reproduces the payoff table and stops. Here is where each one stops.
Setup
Send the term sheet.
A note is a short spec: the payoff as segments, the observation rule, the tenor, the issuer’s rating. Or the Excel a desk already keeps, which the engine reads directly. Priced in about half a second per note; the same engine is reachable over the API for a desk that runs its own tooling.
Next on the bench
Built in the open.
Not shipped yet, listed anyway, so everything above stays exactly what the engine answers today.
The most common structure on the street, refused today rather than priced badly.
The open row: closing the gap between the model’s worst-of prices and issuers’ own disclosed values.
Value, negotiate and book views, so a term sheet can be priced without anyone touching a spec file.
Desk access
The payoff, and what the issuer kept.
For wealth desks, distributors and family offices that place market-linked notes. Send a live term sheet and we return the fair-value band, the retained margin, and the participation you should be asking for. No issuer is named in anything that leaves the desk.
Get access