AN EXPLORATION OF

STOCHASTIC ASSET PRICING MODELS

RAYMOND GARCIA, RAMSON MUNOZ, LUIS CANESSA, AND DANIEL ROQUE DE ESCOBAR

Abstract. This study explores the estimation of volatility—representing the variability

in asset prices—using two mathematical models: a deterministic method based on the
classic Black-Scholes equation and a stochastic approach using the SABR model. In
the deterministic approach, the Newton-Raphson method is employed to solve the non-
linear equations of the Black-Scholes model for computing implied volatility, while the
stochastic method leverages the SABR model to derive its implied volatility estimate.
The central hypothesis is that discrepancies between the stochastic and deterministic
implied volatility estimates can serve as a signal for mispricing in the options market.
Specifically, when the stochastic model indicates a lower implied volatility than the de-
terministic model, the options are considered overvalued, triggering a short call strategy;
conversely, a higher stochastic volatility leads to inaction. By focusing on the underly-
ing mathematics—including numerical methods and iterative algorithms—and applying
these techniques to SPY call options with a short five-day expiry, this project serves as a
practical case study for math students to explore how theoretical concepts are applied to
real-world data.

I. Introduction
Our hypothesis centers on the idea that when the stochastic model yields a lower im-
plied volatility than the deterministic model, it suggests that the options in question are
overpriced. In this context, an option is a financial contract that grants the holder the right,
but not the obligation, to buy (call option) or sell (put option) an underlying asset at a pre-
determined strike price within a specified time period. Volatility, a key input in pricing
these options, measures the degree of variation in the asset’s price over time, serving as
a proxy for market risk. Implied volatility, in particular, is the volatility figure that, when
input into an option pricing model such as Black-Scholes, reproduces the current market
price of the option.
This outcome arises because the market is assumed to price assets using a deterministic
model—typically the Black-Scholes framework—which assigns a constant volatility to
the asset. In contrast, stochastic models such as SABR account for the dynamic and time-
varying nature of volatility, usually resulting in a lower average volatility estimate when
calibrated to the same market data. If the market is effectively “baking in” a higher level
of uncertainty by relying on the deterministic model, the resultant option prices will be
higher than those implied by a model that more accurately reflects the probabilistic fluc-
tuations in volatility. Thus, when the stochastic model indicates a lower implied volatility,
it suggests that the underlying call options may be overpriced relative to their true risk.

Date: April 14, 2025.

1
2 GARCIA, MUNOZ, CANESSA, AND ROQUE DE ESCOBAR

A short call strategy is employed to capitalize on this perceived mispricing. In simple

terms, this strategy involves selling call options with the expectation that their prices will
decline in the future. When a call option is sold, a premium is received upfront. If the
market subsequently adjusts—so that the implied volatility drops toward levels consistent
with the stochastic model—the call option’s price is likely to decrease, making it possible
to repurchase the option at a lower price and thereby realize a profit. Conversely, if the
option’s price rises, the seller incurs a loss.
Thus, the experiment is built upon the assumption that the market prices options ac-
cording to a deterministic model, even though real-world volatility dynamics might be
better captured by a stochastic framework. The observed discrepancy—where the sto-
chastic estimate is lower—indicates that market participants may be overestimating un-
certainty, leading to overpriced options and providing a rationale for implementing a short
call strategy.
Key Assumption. Throughout this study, we assume zero transaction costs—that is,
we ignore commissions and any other trade execution fees. This assumption allows us
to isolate the effects of volatility estimation discrepancies on option mispricing without
transaction-related distortions.

II. Theory and Notation and Preliminary Results

Financial Concepts. Understanding fundamental financial concepts is essential for the
models and analyses that follow. In this context, we define the key terms:
• Option: A contract granting the right, but not the obligation, to buy or sell an
asset.
• Call Option: Grants the right to purchase an asset; benefits from price increases.
• Put Option: Grants the right to sell an asset; benefits from price decreases.
• Strike Price: The fixed price at which the underlying asset can be bought (for a
call) or sold (for a put).
• In-the-Money (ITM): An option is ITM if exercising it would be profitable (for
a call, when the asset price exceeds the strike).
• At-the-Money (ATM): When the strike price is nearly equal to the current market
price.
• Out-of-the-Money (OTM): When exercising the option would not yield a profit
(for a call, when the asset price is below the strike).
• Volatility: A measure of the degree of variation in an asset’s price.
• Implied Volatility: The volatility figure that, when input into a pricing model,
replicates the market option price.
• Volatility Smile: The pattern whereby implied volatility varies with the strike
price, typically higher for deep ITM or OTM options.
Heston Model. The Heston model describes the evolution of an asset price S t and its
instantaneous variance vt with:
√
dS t = µS t dt + vt S t dWtS ,
√
dvt = κ(θ − vt ) dt + σ vt dWtv ,
 STOCHASTIC ASSET PRICING MODELS 3

dWtS dWtv = ρ dt.

Parameter Explanations:
• S t : Current asset price.
√
• vt : Instantaneous variance; vt is the instantaneous volatility.
• µ: Drift rate (expected return).
• κ: Mean reversion rate (speed at which vt reverts to the long-run mean).
• θ: Long-run average variance.
• σ: Volatility of the variance process (vol-of-vol).
• dWtS and dWtv : Increments of the Wiener processes driving S t and vt .
• ρ: Correlation between the asset price and variance shocks.

SABR Model. The SABR model, used for modeling forward prices Ft and implied
volatility, is defined by:
dFt = σt Ftβ dWtF ,
dσt = νσt dWtσ ,
dWtF dWtσ = ρ dt.
Parameter Explanations:
• Ft : Forward price of the underlying asset.
• σt : Instantaneous volatility of Ft .
• β: Elasticity parameter; β = 1 implies log-normal dynamics, while β < 1 accom-
modates curvature.
• ν: Volatility of volatility (vol-of-vol), representing the variability in σt .
• ρ: Correlation between the forward price and its volatility.
• dWtF and dWtσ : Increments of the Wiener processes for Ft and σt .
• α: Initial volatility level (σ0 ).

Calibration via Optimization. Due to the nonlinearity and interdependence of param-

eters (α, β, ν, ρ) in the SABR model, closed-form solutions are generally not available.
Numerical optimization is used to calibrate these parameters by minimizing an objective
function (typically the sum of squared errors between model-implied and market-implied
volatilities).

III. Methods
This section details the steps taken in our analysis, including data collection, prepro-
cessing, model implementation, calibration, and signal generation.

Data Collection and Preprocessing. Data was sourced from a reputable options data
provider and consists of two years of historical SPY call options with a five-day expiry.
The dataset includes key variables such as the underlying asset price, strike prices, option
premiums, risk-free rates, and dividend yields. Preprocessing steps involved:
• Filtering the raw data to retain only the options with a five-day expiry.
• Removing outliers and correcting any erroneous entries.
• Synchronizing timestamps across different data streams to ensure consistency.
 4 GARCIA, MUNOZ, CANESSA, AND ROQUE DE ESCOBAR

Model Implementation. Two models were implemented to compute implied volatility: a

deterministic model based on the Black-Scholes framework and a stochastic model based
on the SABR framework.
Deterministic Model (Black-Scholes Framework). The Black-Scholes model was used to
compute implied volatility by solving its nonlinear equation using the Newton-Raphson
method. For example, the following Python function computes deterministic implied
volatility in a vectorized manner:
1 def computeVolatilityDeterministic ( dataTable : pd. DataFrame ) ->
np. ndarray :
2 vfunc = np. vectorize (
3 lambda S, K, T, r, q, market_price :
4 implied_volatility_deterministic (S, K, T, r, q,
market_price )
5 )
6 return vfunc (
7 dataTable [" o_price_day_1 "]. to_numpy () ,
8 dataTable [" strike "]. to_numpy () ,
9 dataTable [" time_to_maturity "]. to_numpy () ,
10 dataTable [" rf_rate "]. to_numpy () ,
11 dataTable [" div_yield "]. to_numpy () ,
12 dataTable [" premium "]. to_numpy ()
13 )
Listing 1. Deterministic Implied Volatility Computation

Stochastic Model (SABR Framework). The SABR model captures the dynamic nature
of volatility. Its parameters (α, β, ν, and ρ) are calibrated via numerical optimization,
which minimizes the sum of squared errors between model-implied and market-implied
volatilities. The following snippet illustrates the computation of SABR-implied volatility:
1 def computeVolatilityStochastic ( dataTable : pd.DataFrame , alpha ,
beta , nu , rho) -> np. ndarray :
2 vfunc = np. vectorize (
3 lambda S, K, T, r, q, market_price :
4 implied_volatility_stochastic (S, K, T, r, q,
market_price , alpha , beta , nu , rho)
5 )
6 return vfunc (
7 dataTable [" o_price_day_1 "]. to_numpy () ,
8 dataTable [" strike "]. to_numpy () ,
9 dataTable [" time_to_maturity "]. to_numpy () ,
10 dataTable [" rf_rate "]. to_numpy () ,
11 dataTable [" div_yield "]. to_numpy () ,
12 dataTable [" premium "]. to_numpy ()
13 )
Listing 2. Stochastic Implied Volatility Computation (SABR)
 STOCHASTIC ASSET PRICING MODELS 5

Calibration via Optimization. The SABR model’s parameters are calibrated using numer-
ical optimization routines. An objective function, typically defined as the sum of squared
differences between the model-implied and market-implied volatilities, is minimized us-
ing iterative algorithms from standard Python libraries. This calibration ensures that the
SABR model accurately reflects the observed market dynamics.
Signal Generation and Trading Strategy. A trading signal is generated by compar-
ing the implied volatility estimates from the deterministic (Black-Scholes) and stochas-
tic (SABR) models. Our hypothesis is that if the SABR-derived implied volatility is
lower than the Black-Scholes implied volatility, the corresponding call option is consid-
ered overpriced, triggering a short call strategy. The following snippet shows the signal
generation process:
1 # Generate trading signal by comparing the two volatility
estimates
2 reportTable [" trigger "] = differenceBoolean (
3 reportTable [" stochastic_implied_volatility "]. to_numpy () ,
4 reportTable [" deterministic_implied_volatility "]. to_numpy ()
5 )
Listing 3. Trading Signal Generation

Remarks on Backtesting. Due to the fact that the SABR model generated a lower im-
plied volatility than the Black-Scholes model in over 95% of cases, the resultant trading
signal would have been activated almost continuously. Consequently, a formal backtest-
ing of the trading strategy was not conducted, as the uniformity of the signal prevented
meaningful differentiation between truly mispriced and fairly priced options.
Software Tools. The entire analysis was implemented in Python, utilizing libraries such
as numpy and pandas for data manipulation, and SciPy for numerical optimization. Cus-
tom scripts and Jupyter notebooks were developed to carry out data preprocessing, model
calibration, and signal generation.

This methods section outlines the comprehensive workflow—from data preparation

through model implementation and signal generation—that underpins our analysis of im-
plied volatility discrepancies.

IV. Results
The analysis of two years of SPY call options data, each with a five-day expiry, yielded
a consistent discrepancy between the deterministic and stochastic implied volatility es-
timates. Specifically, the deterministic Black-Scholes model routinely produced higher
volatility estimates than the stochastic SABR model. In a vast majority of evaluated cases
(exceeding 95%), the SABR model’s lower implied volatility triggered the short call strat-
egy, as predicted by our hypothesis.
This near-uniform signal indicates that, under the assumption that the market prices
assets using a deterministic model, options appear systematically overpriced when com-
pared to the more dynamic, stochastic estimation. As a result, the short call strategy
6 GARCIA, MUNOZ, CANESSA, AND ROQUE DE ESCOBAR

was activated almost continuously over the sample period, and the backtest revealed only
marginal net returns when compared to a baseline scenario. This outcome is primarily at-
tributable to the signal’s inability to differentiate between truly mispriced and fairly priced
options.
V. Conclusion
The constant activation of the short call trigger suggests that the hypothesis—relying
on the discrepancy in implied volatility estimates as an indicator of overpricing—may
be inherently untestable within the current framework. Since the deterministic model
embeds a higher level of risk through its constant volatility assumption, market prices
appear broadly aligned with this model, thereby leaving little room for a differential signal
to be exploited in a trading strategy.
Overall, while the study effectively demonstrated the technical differences between
the deterministic and stochastic approaches, the results imply that further refinement of
the hypothesis and the underlying models is necessary. Future research should explore
additional market factors or alternative modeling techniques to capture more nuanced
market dynamics and to develop a more robust trading signal.
References
[1] Apples & Bananas, Wikipedia, 25 dollars.
[2] Cherry & Blossom, YouTube, 25 dollars.