Nicolas Liebaert Français

Pricing a Phoenix autocall: a 7% coupon, and most paths end within two quarters

A Phoenix autocall advertises a coupon of 7% a year. On this page you can price one and see how much coupon it pays, how long it lasts and when it loses capital.

99.24 Fair price for 100 invested 95% confidence interval 99.17 to 99.31
3.50 Median coupons collected, in points Mean 7.60, pulled up by the paths that last
84.2% Paths repaid early 48.2% on the first observation date
8.6% Paths that lose capital, at the pricing drift 2.4% at a drift of 7.5% a year

Phoenix autocall, quarterly observation dates

Saved result, 200,000 paths

The live model will load here

The note
The market

Worth 99.24 for 100 invested. At the pricing drift, 84.2% of paths are repaid early and 8.6% lose capital (2.4% at a drift of 7.5%).

When the note ends. Each bar is the probability of repayment on that date. The last bar is the probability that it reaches maturity.
Table of this chart
Observation date (years)Probability
Value of the note against the index level. The barriers stay where they were set on day one.
Table of this chart
SpotValue of the note
Price
99.24
95% interval 99.17 to 99.31
Expected life
1.38
years, on average
Repaid early
84.2%
repaid at 100 before maturity
Capital loss
8.6%
at the pricing drift; 2.4% at a drift of 7.5%
Average repayment after a loss
44.9
out of 100, coupons not counted
Worst repayment
11.7
out of 100, on any path
Coupons collected, mean
7.60
points of nominal
Coupons collected, median
3.50
points of nominal
Delta
0.20
± 0.01, per 1% of spot
Gamma
-0.03
± 0.03, per 1% of spot
Vega
-0.45
± 0.01, per volatility point
Rho
-0.005
± 0.002, per basis point (0.01 percentage point)
Amounts are in points of nominal: 100 means the capital is repaid in full. The greeks show how the price reacts: delta and gamma to the spot, vega to volatility, rho to the interest rate. The ± beside each one is its 95% confidence interval. Greeks from a Monte Carlo are noisy: at 20,000 paths gamma is mostly noise, and High precision divides each interval by about three. Even at 200,000 paths its interval includes zero. They update when you release a slider, because they need four extra pricings. The paths are antithetic: each random path is paired with its mirror image, which reduces simulation noise. The seed (the start of the random sequence) is fixed, so the same settings always give the same numbers. Python, NumPy and SciPy run in your browser, in a background thread, so you can keep using the page. The first download is about 31 MB, from this site. Your browser then keeps the files.

How the note works

Every three months, on an observation date, the note compares the stock index with its starting level.

If the index is above the coupon barrier, the note pays a coupon. A missed coupon is paid later, together with the current one, the next time the index is above that barrier. The market calls this a memory coupon.

If the index is above the autocall barrier, the note is repaid at 100 with that day’s coupon, and it ends. This note can be repaid from the first observation date. Many real notes cannot be repaid in their first year, which would change the 48% above.

If the note reaches maturity, the investor gets 100 back as long as the index is above the capital barrier. If the index is below it, the investor loses the same percentage as the index.

This last condition pays for the coupon. In effect, the investor has sold the bank a put on the index (an option that pays off when the index falls). The coupon is that premium, added to the interest and the dividends the investor gives up.

What to try in the demo

Drag the volatility up. The price falls, because the investor is short volatility. The note contains a down-and-in put (a put that only counts if the index ends below the capital barrier) sold to the bank, which is worth more when volatility rises, while the coupon stays fixed. At the default settings the vega is about -0.45 per volatility point.

Drag the spot down towards 60. The value falls faster and faster. Between a spot of 100 and 80, the note loses about 0.4 points of value for each point of spot. Between 70 and 60, it loses about 1.2. Between the coupon barrier and the capital barrier, the note pays no coupon and can still lose capital.

Compare the mean and the median coupon. At the default settings the mean is about 7.6 points and the median about 3.5. The 16% of paths that reach maturity keep paying for years, and they pull the mean up. A term sheet (the summary of the contract) quotes the coupon rate. It does not show this gap.

Limits of this model

No bank risk in the price. The model discounts every payment at the risk-free rate. A real note is priced with the funding cost of the issuing bank, and the investor bears the risk that the bank fails. Both lower the value. So the fair price shown here is higher than the one a bank would quote.

One flat volatility. The model uses one volatility, one interest rate and one dividend yield. This shows the shape of the payoff, but a trading desk would not price with it. In real markets, puts with strikes (exercise prices) far below today’s level cost more, and the capital barrier sits exactly there. A flat volatility makes that risk too cheap, so the note looks better than it is. The options page shows this effect on a real option chain.

Risk-neutral probabilities. The probabilities come from the pricing model and do not use history. In the model the index drifts at the interest rate minus the dividend yield, about 0.5% a year. Real indices have usually drifted up faster. With a drift of 2% a year, about 87% are repaid early and 6.8% lose capital. With 7.5%, a generous case, about 93% and 2.4%.

One index. Many real notes are linked to a basket of stocks and pay according to the worst one (a worst-of note). That is riskier, and this model cannot show it.

The full source is at github.com/Nicolas8330/structured-products-pricer. Its 16 tests include checks of the pricing engine against closed-form prices (exact formulas).