Author

Clement D.

Best Quant Cities
Jobs

Top Cities for High-Paying C++ Quantitative Roles

by Clement D.

If you’re a skilled C++ developer with an interest in high-performance finance, the world of quantitative trading offers some of the most lucrative roles available today. These positions, often found at hedge funds, proprietary trading firms, and investment banks, demand a blend of low-latency programming expertise, mathematical insight, and real-time systems knowledge. While demand exists globally, a handful of cities stand out for consistently offering the highest compensation packages.

From Wall Street to Canary Wharf, certain global financial hubs continue to dominate the quant talent market. These cities not only house the worldโ€™s top funds and trading desks but also offer competitive salaries, generous bonuses, and exposure to cutting-edge infrastructure. In this article, we break down the top five cities for high-paying C++ quantitative roles, supported by up-to-date salary data and market trends.

1. New York City, USA

Hereโ€™s a detailed look at New York City, the top destination globally for high-paying C++ quantitative developer roles:

๐Ÿ’ฐ Salary Ranges & Market Averages

๐Ÿข Top Firms & Roles

๐Ÿ” 3. Role Levels & Progression

๐Ÿง  4. Compensation Structure

  • Compensation is a blend of base + bonus + potential stock/equity. Bonuses can equal or exceed base salaries, especially in top-tier firms .
  • Reddit users note: โ€œQuant Developers: $300kโ€“1M โ€ฆ top graduates โ€ฆ guarantees in excess of $500k for the 1st yearโ€ at elite shops reddit.com+1efinancialcareers.com+1.

๐Ÿ“ 5. Role Types & Relevance of C++

  • Most high-paying roles are C++-centric, especially in low-latency trading, execution systems, and infrastructure for quant analytics.
  • Job listings for C++ Low Latency Trading Systems Dev show base pay typically $150kโ€“$300k, with bonus upsides indeed.com+3linkedin.com+3bloomberg.avature.net+3.

๐Ÿงฉ ๐Ÿ‘‰ Summary Snapshot

Role LevelBase PayTotal Compensation
Entryโ€“Mid (Hedge/Fund)$150kโ€“$225k$225kโ€“$350k (with bonus)
Senior/Elite$250kโ€“$350k+$400kโ€“$700k+ (with bonus/equity)
Average (mid-career)$326k base$376k total comp

Bottom line:
New York City remains the gold standard for C++ quantitative developers. At top-tier firms, you’ll see base salaries of $250k+ and total compensation reaching $500kโ€“$700k, especially at hedge funds and prop trading shops. Even outside the elite circles, mid-tier roles offer $150kโ€“$200k base with solid bonus structures. C++ expertise in low-latency systems is an exceptionally valued skill in this market.

2. San Francisco Bay Area / Silicon Valley, USA

Hereโ€™s a detailed exploration of San Francisco Bay Area / Silicon Valley, demonstrating why it ranks as one of the top-paying regions for C++ quantitative developers:

๐Ÿ’ผ Salary Overview for Quantitative Developers

๐Ÿ“Š Comparison with California & National Averages

๐Ÿ› ๏ธ C++ Developer Base Salaries

  • Indeed reports average base pay for general C++ Developers in San Francisco at $177,272/year, with a range from $123k to $254k salary.com+5indeed.com+5indeed.com+5.
  • Salary.com notes senior C++ Developer base ranges from $177,904 to $212,377, averaging $193,584 salary.com.

๐Ÿš€ Silicon Valley Premium

  • C++ developer salaries in the broader Bay Area average $162k, with wide dispersionโ€”from $95k to $460kโ€”reflecting startup equity upside wellfound.com.
  • San Jose (next to SF) offers higher comps for Quant Dev roles: total compensation averages $429,654/year, with base around $225,027/year wellfound.com+3glassdoor.com+3ziprecruiter.com+3.

๐Ÿ’ต Total Compensation Breakdown

Role TypeBase Pay RangeTotal Comp Range
Quant Dev (SF)$131kโ€“$173k$248kโ€“$390k+
General C++ Dev (SF)$123kโ€“$212kโ€”
Quant Dev (San Jose)~$225k (base)~$430k total avg

โš™๏ธ Role Types & C++ Relevance

High-paying roles typically emphasize low-latency C++ engineering within algorithmic trading engines, pricing libraries, and high-performance analytics platforms.

  • Compensation structure includes base + cash bonus + sometimes stock/equity, especially at startups and tech-influenced trading shops.

๐ŸŽฏ Bottom Line

  • Base pay for Bay Area quant C++ roles ranges $130kโ€“$225k+, depending on seniority and location.
  • Total compensation regularly reaches $250kโ€“$400k in San Francisco, with $430k+ in San Jose, especially at elite or startup-oriented firms.

3. London, UK

๐Ÿ’ท Salary Range โ€“ Base & Total Compensation

๐Ÿ“ˆ Premium Compensation for C++ & HFT Roles

๐ŸŽฏ Senior & Front-Office Engineer Earnings

  • Roles in hedge funds and high-frequency trading often target senior candidates with ยฃ150kโ€“ยฃ350k total compensation, emphasizing front-office C++ expertise .
  • Client Server listings feature C++/Python Quant Dev roles with ยฃ110kโ€“ยฃ175k base, plus potential bonuses worth multiple base salaries clientserver.com.

๐Ÿงฉ Skill Demand & Market Pressure

  • ITJobsWatch data shows the UK median for Quant Developer roles is ยฃ140k, with Londonโ€™s median at ยฃ150k morganmckinley.com+7itjobswatch.co.uk+7oxfordknight.co.uk+7.
  • C++ and low-latency expertise feature prominently in job ads (~60% mention C++), especially within hedge funds and algorithmic trading shops .

โš–๏ธ Junior to Senior Progression Path

Experience LevelBase PayTotal Compensation
Entry (0โ€“3 yrs)ยฃ65kโ€“ยฃ100kยฃ70kโ€“ยฃ125k
Mid (3โ€“5 yrs)ยฃ105kโ€“ยฃ150kยฃ150kโ€“ยฃ200k
Senior (5+ yrs)ยฃ150kโ€“ยฃ200k+ยฃ200kโ€“ยฃ350k+

๐Ÿงญ Why London Holds Its Ground

  • As a major global financial hub, London hosts numerous hedge funds, trading desks, and investment banks that depend on low-latency C++ infrastructure .
  • Market demand remains strong, with job vacancies growing and salaries up ~9% year-on-year, according to ITJobsWatch .

โœ… Summary Insight

London offers compelling opportunities for C++ quantitative developers:

  • Base salaries typically: ยฃ65kโ€“ยฃ150k, rising with seniority.
  • Total compensation often ranges from ยฃ100k to ยฃ350k+ at elite firms.
  • Top-tier roles at HFT/hedge funds pay aggressively, reflecting C++โ€™s strategic value in low-latency systems.
0 comments
C++ quant jobs
Jobs

C++ Quantitative Developers: A Skyrocketing Job Market

by Clement D.

The job market for C++ quantitative developers is experiencing a major surge. Driven by the relentless demand for low-latency execution, many hedge funds and trading firms are ramping up hiring. C++ remains the gold standard for performance-critical systems in finance โ€” and its dominance is growing.
Top firms like Citadel, Jane Street, and Jump Trading are offering eye-watering compensation packages to attract talent.

In London, New York, and Singapore, six-figure base salaries are now entry-level โ€” with total comp often exceeding $500k.
Real-time risk, high-frequency trading, and exotic derivatives desks all need C++ expertise.
The rise of data-driven modeling has only reinforced the need for tight integration between quant models and execution engines.
Firms want devs who can code, optimize, and understand the math โ€” and C++ sits at that intersection.
With competition heating up, even junior roles now demand systems-level thinking and modern C++ fluency.
For quants and devs alike, now is a golden moment to ride the C++ finance wave.

1. Some Data about Salaries

The demand for C++ Quant Developers in the UK has exploded โ€” and salaries reflect it. The median salary has jumped to ยฃ170,000, up 17.24% year-on-year, continuing a multi-year upward trend. Even more striking, the 10th percentile salary has more than doubled since 2024, rising from ยฃ56,250 to ยฃ135,000, suggesting that entry-level roles are commanding mid-career paychecks.

The number of permanent roles has also nearly doubled in a year, with 33 positions advertised versus just 18 the year prior. C++ quant roles now represent 0.058% of all UK job ads, a 3.6x increase in relative share โ€” highlighting how niche, high-impact, and in-demand this skillset has become.

Whether you’re a seasoned systems programmer or a mathematically-minded developer eyeing the finance sector, the numbers don’t lie: it’s a C++ quant boom.

2. The Trend is Volatile but Clearly Goes Up

This 20-year salary trend graph shows a clear and accelerating rise in compensation for C++ Quantitative Developers in the UK. After a decade of relative stability from 2005 to 2015, salaries began a marked upward shift around 2018, aligning with the growing demand for low-latency systems and tighter model-to-execution integration. Since 2020, volatility has increased โ€” but so has the upside. The median salary line (orange) now sits firmly above ยฃ150,000, with the top 10% breaching the ยฃ200,000+ mark. Even the 25th percentile has climbed significantly, pointing to strong tailwinds across all seniority levels. As of mid-2025, the trend is steeply upward โ€” a reflection of how C++ has reasserted itself as a core technology in quant finance.

3. The Tools, Skills, Exposure and Libraries Required

The C++ Quant Developer role is no longer just about knowing C++. According to data from the six months leading up to June 2025, Low Latency (93.94%), Equities, and Hedge Funds appear alongside C++ in nearly all job ads, signaling a strong demand for developers who understand real-time execution environments in capital markets.

Interestingly, Python (90.91%) appears just as frequently โ€” reinforcing the industry’s shift toward hybrid devs who can optimize in C++ and prototype in Python.
Skills like Linux, Multithreading, Algorithms, and Data Structures remain core, while mentions of Boost, Test Automation, and Order Management show that system reliability and front-office tooling are just as valued as raw performance.

Whether itโ€™s Quantitative Trading, Market Making, or Greenfield Projects, this skill map clearly shows that modern quant devs must span systems engineering, financial markets, and rapid prototyping โ€” a rare and highly-paid blend.

4. Conclusion

The data is unambiguous: C++ Quantitative Developers are among the most sought-after professionals in finance today. Salaries are surging, with median comp hitting ยฃ170,000, and top roles exceeding ยฃ200,000+. Even the lowest percentiles are rising fast โ€” entry-level is no longer โ€œjuniorโ€ in pay. Demand has more than doubled year-on-year, with job postings climbing steadily. C++ sits at the core of high-frequency trading, real-time risk, and complex derivatives pricing.

But employers arenโ€™t just hiring C++ coders โ€” they want versatile technologists. Fluency in Python, Linux, Multithreading, and Low Latency architecture is essential. Firms want devs who can design systems, optimize them, and understand market dynamics. This is no longer a back-office role โ€” the Front Office is calling, and it pays.

Whether youโ€™re a quant with systems chops or a dev learning finance, now is the time to strike. The most competitive candidates understand both algorithms and alpha. They build fast, test fast, and deploy into production with confidence. The C++ quant role is evolving โ€” itโ€™s becoming broader, better-paid, and more central to business.

This is not a bubble. Itโ€™s a structural shift. As market infrastructure becomes more automated and data-hungry, firms will invest heavily in top-tier engineering. That means a long runway for anyone investing in the right skills now.
If youโ€™ve been on the fence about switching into finance โ€” consider this your signal. And if you’re already here: it’s time to double down on your edge.

0 comments
moving average interview
Interview

Calculate Moving Average in C++ in O(1) โ€“ An Interview-Style Problem

by Clement D.

A classic interview question in quantitative finance or software engineering roles is:
“Design a data structure that calculates the moving average of a stock price stream in O(1) time per update.”.
How to calculate the moving average in C++ in an optimal way?

Letโ€™s tackle this with a focus on Microsoft (MSFT) stock, although the solution is applicable to any time-series financial instrument. We’ll use C++ to build an efficient, clean implementation suitable for production-grade quant systems.

1. Problem Statement

Design a class that efficiently calculates the moving average of the last N stock prices.

Your class should support:

  • void addPrice(double price): Adds the latest price.
  • double getAverage(): Returns the average of the last N prices.

Constraints:

  • The moving average must be updated in O(1) time per price.
  • Handle the case where fewer than N prices have been added.

Implement this in C++.

You will need to complete the following code:

#include <vector>


class MovingAverage {
public:
    explicit MovingAverage(int size);
    void addPrice(double price);
    double getAverage() const;
private:
    std::vector<double> buffer;
    int maxSize;
    double sum = 0.0;
};

Imagine the following historical prices for Microsoft stocks:

DayPrice (USD)
1400.0
2402.5
3405.0
4410.0
5412.0
6415.5

And assume we want to calculate the moving average of prices on 3 days, everyday, as a test.

2. A Naive Implementation in O(N)

Let’s start with a first implementation using `std::accumulate`.

It’s defined as follow in the C++ documentation:
“std::accumulate Computes the sum of the given value init and the elements in the range [first, last)“. The last iterator is not included in the operation. This is a half-open interval, written as.

std::accumulate is O(N) in time complexity.

Let’s use it to calculate the MSTF stock price moving average in C++:

#include <iostream>
#include <vector>
#include <numeric>

class MovingAverage {
public:
    explicit MovingAverage(int size) : maxSize(size) {}

    void addPrice(double price) {
        buffer.push_back(price);
        if (buffer.size() > maxSize) {
            buffer.erase(buffer.begin()); // O(N)
        }
    }

    double getAverage() const {
        if (buffer.empty()) return 0.0;
        double sum = std::accumulate(buffer.begin(), buffer.end(), 0.0); // O(N)
        return sum / buffer.size();
    }

private:
    std::vector<double> buffer;
    int maxSize;
};

int main() {
    MovingAverage ma(3); // 3-day moving average
    std::vector<double> msftPrices = {400.0, 402.5, 405.0, 410.0, 412.0, 415.5};

    for (size_t i = 0; i < msftPrices.size(); ++i) {
        ma.addPrice(msftPrices[i]);
        std::cout << "Day " << i + 1 << " - Price: " << msftPrices[i]
                  << ", 3-day MA: " << ma.getAverage() << std::endl;
    }

    return 0;
}

Every new day the moving average is re-calculated from scratch, not ideal! But it gives the right results:

โžœ  build ./movingaverage 
Day 1 - Price: 400, 3-day MA: 400
Day 2 - Price: 402.5, 3-day MA: 401.25
Day 3 - Price: 405, 3-day MA: 402.5
Day 4 - Price: 410, 3-day MA: 405.833
Day 5 - Price: 412, 3-day MA: 409
Day 6 - Price: 415.5, 3-day MA: 412.5

3. An optimal Implementation in O(1)

Let’s now do it in an iterative way and just add the value to the former sum to calculate a performant moving average in C++:

#include <iostream>
#include <vector>

class MovingAverage {
public:
    explicit MovingAverage(int size)
        : buffer(size, 0.0), maxSize(size) {}

    void addPrice(double price) {
        sum -= buffer[index];        // Subtract the value being overwritten
        sum += price;                // Add the new value
        buffer[index] = price;       // Overwrite the old value
        index = (index + 1) % maxSize;

        if (count < maxSize) count++;
    }

    double getAverage() const {
        return count == 0 ? 0.0 : sum / count;
    }

private:
    std::vector<double> buffer;
    int maxSize;
    int index = 0;
    int count = 0;
    double sum = 0.0;
};

int main() {
    // Example: 3-day moving average for Microsoft stock prices
    MovingAverage ma(3);
    std::vector<double> msftPrices = {400.0, 402.5, 405.0, 410.0, 412.0, 415.5};

    for (size_t i = 0; i < msftPrices.size(); ++i) {
        ma.addPrice(msftPrices[i]);
        std::cout << "Day " << i + 1
                  << " - Price: " << msftPrices[i]
                  << ", 3-day MA: " << ma.getAverage()
                  << std::endl;
    }

    return 0;
}

This time, no re-calculation, each time a new price gets in, we add it and average it on the fly. Although it looks better, the vector data structure in that case is not ideal.

Another approach is to use a double-entry queue: std::deque.

It perfectly suits the sliding window use case as we can pop and add elements both sides of the data structure in a very easy way (push_back/pop_front):

#include <iostream>
#include <deque>

/**
 * MovingAverage maintains a fixed-size sliding window of the most recent prices
 * and efficiently computes their average.
 */
class MovingAverage {
public:
    explicit MovingAverage(int windowSize) : size_(windowSize), sum_(0.0) {}

    // Adds a new price to the window and updates the sum accordingly
    void addPrice(double price) {
        prices_.push_back(price);
        sum_ += price;

        // Remove the oldest price if the window is too big
        if (prices_.size() > size_) {
            sum_ -= prices_.front();
            prices_.pop_front();
        }
    }

    // Returns the current moving average
    double getMovingAverage() const {
        if (prices_.empty()) return 0.0;
        return sum_ / prices_.size();
    }

private:
    std::deque<double> prices_;
    double sum_;
    int size_;
};

int main() {
    MovingAverage ma(3); // 3-day moving average

    // Historical Microsoft stock prices
    std::vector<double> msftPrices = {400.0, 402.5, 405.0, 410.0, 412.0, 415.5};

    std::cout << "Microsoft 3-day Moving Average Calculation:\n" << std::endl;

    for (size_t i = 0; i < msftPrices.size(); ++i) {
        ma.addPrice(msftPrices[i]);
        std::cout << "Day " << i + 1 << " - Price: " << msftPrices[i]
                  << ", 3-day MA: " << ma.getMovingAverage() << std::endl;
    }

    return 0;
}


It only adds/pops elements, then updates the sum, then the average.

In other terms: the time complexity O(1).

Let’s run it:

โžœ  build ./movingaverage
Day 1 - Price: 400, 3-day MA: 400
Day 2 - Price: 402.5, 3-day MA: 401.25
Day 3 - Price: 405, 3-day MA: 402.5
Day 4 - Price: 410, 3-day MA: 405.833
Day 5 - Price: 412, 3-day MA: 409
Day 6 - Price: 415.5, 3-day MA: 412.5

The same results but way faster!

And you pass your interview to calculate the moving average in C++ in the most performant way.

4. Common STL Container Member Functions & Tips (with Vector vs Deque)

โœ… Applies to Both std::vector<T> and std::deque<T>

SyntaxDescriptionNotes / Gotchas
container.begin()Iterator to first elementUsed in range-based loops and STL algorithms ([begin, end))
container.end()Iterator past the last elementNon-inclusive range end
container.size()Number of elementsType is size_t โ€” beware of unsigned vs signed int bugs
container.empty()true if container is emptySafer than size() == 0
container.front()First elementโš ๏ธ Undefined if container is empty
container.back()Last elementโš ๏ธ Undefined if container is empty
container.push_back(x)Add x to the endO(1) amortized for vector, always O(1) for deque
container.pop_back()Remove last elementโš ๏ธ Undefined if empty
std::accumulate(begin, end, init)Sum or fold values over rangeFrom <numeric> โ€” works on both vector and deque

๐ŸŸฆ std::deque<T> Specific

SyntaxDescriptionNotes / Gotchas
container.push_front(x)Add x to frontO(1); ideal for queues and sliding windows
container.pop_front()Remove first elementO(1); โš ๏ธ Undefined if empty
container.erase(it)Remove element at iteratorO(1) when removing from front/back
โœ… Random access ([i])SupportedSlightly slower than vector (more overhead under the hood)
0 comments
yield curve
Bonds

Building a Yield Curve in C++: Theory and Implementation

by Clement D.

In fixed income markets, the yield curve is a fundamental tool that maps interest rates to different maturities. It underpins the pricing of bonds, swaps, and other financial instruments. How to build a yield curve in C++?

Traders, quants, and risk managers rely on it daily to discount future cash flows. This article explores how to build a basic zero-coupon yield curve using C++, starting from market data and ending with interpolated discount factors. We’ll also compare our results with QuantLib, the industry-standard quantitative finance library.

1. What’s a Yield Curve?

At its core, the yield curve captures how interest rates change with the length of time you lend money. Hereโ€™s an example of hypothetical yields across maturities:

MaturityYield (%)
Overnight (ON)1.00
1 Month1.20
3 Months1.35
6 Months1.50
1 Year1.80
2 Years2.10
5 Years2.60
10 Years3.00
30 Years3.20

Plotted on a graph, these points form the yield curve โ€” typically upward sloping, as longer maturities usually command higher yields due to risk and time value of money.

Regular Yield Curve

However, in stressed environments, the curve can flatten or even invert (e.g., 2-year yield > 10-year yield), often signaling economic uncertainty:

Inverted Yield Curve

Here is the plot of a hypothetical inverted yield curve, where shorter-term yields are higher than longer-term ones. This often reflects market expectations of economic slowdown or interest rate cuts in the future.

When investors expect future economic weakness, they begin to anticipate interest rate cuts by the central bank. As a result:

  • Demand increases for long-term bonds (seen as safe havens), which pushes their prices up and their yields down.
  • Meanwhile, short-term rates may remain high due to current central bank policy (e.g. inflation control).

This flips the curve: short-term yields become higher than long-term yields.

2. Key Concepts Behind Yield Curve Construction

Before jumping into code, itโ€™s important to understand how yield curves are built from market data. In practice, we donโ€™t observe a complete yield curve directly, we build (or โ€œbootstrapโ€) it from liquid instruments such as:

  • Deposits (e.g., overnight to 6 months)
  • Forward Rate Agreements (FRAs) and Futures
  • Swaps (e.g., 1-year to 30-year)

These instruments give us information about specific points on the curve. QuantLib uses these to construct a continuous curve using interpolation (e.g., linear, log-linear) between observed data points.

Some essential building blocks in QuantLib include:

  • RateHelpers: Abstractions that turn market quotes (like deposit or swap rates) into bootstrapping constraints.
  • DayCount conventions and Calendars: Needed for accurate date and interest calculations.
  • YieldTermStructure: The central object representing the term structure of interest rates, which you can query for zero rates, discount factors, or forward rates.

QuantLib lets you define all of this in a modular way, so you can plug in market data and generate an accurate, arbitrage-free curve for pricing and risk analysis.

3. Implement in C++ with Quantlib

To construct a yield curve in QuantLib, you’ll typically bootstrap it from a mix of deposit rates and swaps. QuantLib provides a flexible interface to handle real-world instruments and interpolate the curve. Below is a minimal C++ example that constructs a USD zero curve from deposit and swap rates:

#include <ql/quantlib.hpp>
#include <iostream>

using namespace QuantLib;

int main() {
    Calendar calendar = UnitedStates(UnitedStates::GovernmentBond);
    Date today(25, June, 2025);
    Settings::instance().evaluationDate() = today;

    // Market quotes
    Rate depositRate = 0.015; // 1.5% for 3M deposit
    Rate swapRate5Y = 0.025;  // 2.5% for 5Y swap

    // Quote handles
    Handle<Quote> dRate(boost::make_shared<SimpleQuote>(depositRate));
    Handle<Quote> sRate(boost::make_shared<SimpleQuote>(swapRate5Y));

    // Day counters and conventions
    DayCounter depositDayCount = Actual360();
    DayCounter curveDayCount = Actual365Fixed();
    Thirty360 swapDayCount(Thirty360::BondBasis);

    // Instrument helpers
    std::vector<boost::shared_ptr<RateHelper>> instruments;

    instruments.push_back(boost::make_shared<DepositRateHelper>(
        dRate, 3 * Months, 2, calendar, ModifiedFollowing, false, depositDayCount));

    instruments.push_back(boost::make_shared<SwapRateHelper>(
        sRate, 5 * Years, calendar, Annual, Unadjusted, swapDayCount,
        boost::make_shared<Euribor6M>()));

    // Construct the yield curve
    boost::shared_ptr<YieldTermStructure> yieldCurve =
        boost::make_shared<PiecewiseYieldCurve<ZeroYield, Linear>>(
            today, instruments, curveDayCount);

    // Example output: discount factor at 2Y
    Date maturity = calendar.advance(today, 2, Years);
    std::cout << "==== Yield Curve ====" << std::endl;
    for (int y = 1; y <= 5; ++y) {
    Date maturity = calendar.advance(today, y, Years);
    double discount = yieldCurve->discount(maturity);
    double zeroRate = yieldCurve->zeroRate(maturity, curveDayCount, Compounded, Annual).rate();

    std::cout << "Maturity: " << y << "Y\t"
              << "Yield: " << std::fixed << std::setprecision(2) << 100 * zeroRate << "%\t"
              << "Discount: " << std::setprecision(5) << discount
              << std::endl;
    }

    return 0;
}

This snippet shows how to:

  • Set up todayโ€™s date and calendar,
  • Define deposit and swap instruments as bootstrapping inputs,
  • Build a PiecewiseYieldCurve using QuantLibโ€™s helper classes,
  • Query the curve for a discount factor.

You can easily extend this with more instruments (FRA, futures, longer swaps) for a realistic curve.

4. Compile, Execute and Plot

After installing quantlib, I compile my code with the following CMakeLists.txt:

cmake_minimum_required(VERSION 3.10)
project(QuantLibImpliedVolExample)

set(CMAKE_CXX_STANDARD 17)

find_package(PkgConfig REQUIRED)
pkg_check_modules(QUANTLIB REQUIRED QuantLib)

include_directories(${QUANTLIB_INCLUDE_DIRS})
link_directories(${QUANTLIB_LIBRARY_DIRS})

add_executable(yieldcurve ../yieldcurve.cpp)
target_link_libraries(yieldcurve ${QUANTLIB_LIBRARIES})

I run:

mkdir build
cd build
cmake ..
make

And run:

โžœ build ./yieldcurve
==== Yield Curve ====
Maturity: 1Y Yield: 1.68% Discount: 0.98345
Maturity: 2Y Yield: 1.89% Discount: 0.96324
Maturity: 3Y Yield: 2.10% Discount: 0.93946
Maturity: 4Y Yield: 2.31% Discount: 0.91272
Maturity: 5Y Yield: 2.52% Discount: 0.88306

And we can plot it too:

So this is the plot of the zero yield curve from 1 to 5 years, based on your QuantLib output. It shows a smooth, gently upward-sloping curve: typical for a healthy interest rate environment.

But why is it so… Linear?

This is due to both the limited number of inputs and the interpolation method applied.

In the example above, the curve is built using QuantLibโ€™s PiecewiseYieldCurve<ZeroYield, Linear>, which performs linear interpolation between the zero rates of the provided instruments. With only a short-term and a long-term point, this leads to a straight-line interpolation between the two, hence the linear shape.

In reality, yield curves typically exhibit curvature:

  • They rise steeply in the short term,
  • Then gradually flatten as maturity increases.

This reflects the market’s expectations about interest rates, inflation, and economic growth over time.

To better approximate real-world behavior, the curve can be constructed using:

  • A richer set of instruments (e.g., deposits, futures, swaps across many maturities),
  • More appropriate interpolation techniques such as LogLinear or Cubic.

5. A More Realistic Curve

This version uses LogLinear interpolation, which interpolates on the log of the discount factors. It results in a more natural term structure than simple linear interpolation, even with only two instruments.

For a truly realistic curve, additional instruments should be incorporated, especially swaps with intermediate maturities:

#include <ql/quantlib.hpp>
#include <iostream>
#include <iomanip>

using namespace QuantLib;

int main() {
    Calendar calendar = UnitedStates(UnitedStates::GovernmentBond);
    Date today(25, June, 2025);
    Settings::instance().evaluationDate() = today;

    // Market quotes
    Rate depositRate = 0.015;    // 1.5% for 3M deposit
    Rate swapRate2Y = 0.020;     // 2.0% for 2Y swap
    Rate swapRate5Y = 0.025;     // 2.5% for 5Y swap
    Rate swapRate10Y = 0.030;    // 3.0% for 10Y swap

    // Quote handles
    Handle<Quote> dRate(boost::make_shared<SimpleQuote>(depositRate));
    Handle<Quote> sRate2Y(boost::make_shared<SimpleQuote>(swapRate2Y));
    Handle<Quote> sRate5Y(boost::make_shared<SimpleQuote>(swapRate5Y));
    Handle<Quote> sRate10Y(boost::make_shared<SimpleQuote>(swapRate10Y));

    // Day counters and conventions
    DayCounter depositDayCount = Actual360();
    DayCounter curveDayCount = Actual365Fixed();
    Thirty360 swapDayCount(Thirty360::BondBasis);

    // Instrument helpers
    std::vector<boost::shared_ptr<RateHelper>> instruments;

    instruments.push_back(boost::make_shared<DepositRateHelper>(
        dRate, 3 * Months, 2, calendar, ModifiedFollowing, false, depositDayCount));

    instruments.push_back(boost::make_shared<SwapRateHelper>(
        sRate2Y, 2 * Years, calendar, Annual, Unadjusted, swapDayCount,
        boost::make_shared<Euribor6M>()));

    instruments.push_back(boost::make_shared<SwapRateHelper>(
        sRate5Y, 5 * Years, calendar, Annual, Unadjusted, swapDayCount,
        boost::make_shared<Euribor6M>()));

    instruments.push_back(boost::make_shared<SwapRateHelper>(
        sRate10Y, 10 * Years, calendar, Annual, Unadjusted, swapDayCount,
        boost::make_shared<Euribor6M>()));

    // Construct the yield curve with Cubic interpolation
    boost::shared_ptr<YieldTermStructure> yieldCurve =
        boost::make_shared<PiecewiseYieldCurve<ZeroYield, Cubic>>(
            today, instruments, curveDayCount);

    // Output: yield curve at each year from 1Y to 10Y
    std::cout << "==== Yield Curve ====" << std::endl;
    for (int y = 1; y <= 10; ++y) {
        Date maturity = calendar.advance(today, y, Years);
        double discount = yieldCurve->discount(maturity);
        double zeroRate = yieldCurve->zeroRate(maturity, curveDayCount, Compounded, Annual).rate();

        std::cout << "Maturity: " << y << "Y\t"
                  << "Yield: " << std::fixed << std::setprecision(2) << 100 * zeroRate << "%\t"
                  << "Discount: " << std::setprecision(5) << discount
                  << std::endl;
    }

    return 0;
}

And let’s run it again:

โžœ  build ./yieldcurve
==== Yield Curve ====
Maturity: 1Y    Yield: 1.67%    Discount: 0.98355
Maturity: 2Y    Yield: 2.00%    Discount: 0.96122
Maturity: 3Y    Yield: 2.20%    Discount: 0.93680
Maturity: 4Y    Yield: 2.37%    Discount: 0.91058
Maturity: 5Y    Yield: 2.51%    Discount: 0.88320
Maturity: 6Y    Yield: 2.64%    Discount: 0.85524
Maturity: 7Y    Yield: 2.75%    Discount: 0.82674
Maturity: 8Y    Yield: 2.86%    Discount: 0.79790
Maturity: 9Y    Yield: 2.96%    Discount: 0.76912
Maturity: 10Y   Yield: 3.05%    Discount: 0.74018

Which gives a better looking yield curve:

0 comments
C++ containers
Data Structures

C++ Sequential Containers for Quants: Speed and Memory

by Clement D.

Sequential containers in C++ store elements in a linear order, where each element is positioned relative to the one before or after it. They’re designed for efficient iteration, fast access, and are typically used when element order matters, such as in time-based data.

The most commonly used sequential containers in the STL are:

  • std::vector: dynamic array with contiguous memory
  • std::deque: double-ended queue with fast insertions/removals at both ends
  • std::array: fixed-size array on the stack
  • std::list / std::forward_list: linked lists

In quantitative finance, sequential containers are often at the heart of systems that process tick data, historical time series, price paths, and rolling windows. When processing millions of data points per day, the choice of container directly impacts latency, memory usage, and cache performance.

1. std::vector: The Workhorse

std::vector is a dynamic array container in C++. It’s one of the most performance-critical tools in a quant developerโ€™s toolbox โ€” especially for time-series data, simulation paths, and numerical computations.

๐Ÿง  Stack vs Heap: The Memory Model

In C++, memory is managed in two primary regions:

Memory TypeDescriptionCharacteristics
StackAutomatically managed, fast, smallFunction-local variables, fixed size
HeapManually or dynamically managedUsed for dynamic memory like new, malloc, or STL containers

๐Ÿ”ธ A std::vector behaves as follows:

  • The vector object itself (pointer, size, capacity) lives on the stack.
  • The elements it stores live in a dynamically allocated heap buffer.

๐Ÿ’ก Consequences:

  • You can create a std::vector of size 1 million without stack overflow.
  • But heap allocation is slower than stack and can cause fragmentation if misused.
  • When passed by value, only the stack-held metadata is copied (until you access elements).

๐Ÿ“ฆ Contiguous Memory Layout

A std::vector stores all its elements in a single, contiguous block of memory on the heap โ€” just like a C-style array.

Why this matters:

  • CPU cache efficiency: Memory is prefetched in blocks โ€” sequential access is blazing fast.
  • Enables pointer arithmetic, SIMD (Single Instruction Multiple Data), and direct interfacing with C APIs.

๐Ÿ’ฅ For quants: When you’re processing millions of floats/doubles (e.g., price histories, simulation paths), contiguous memory layout can be a 10ร— performance multiplier vs pointer-chasing containers like std::list.

โฑ Time Complexity

OperationTime ComplexityNotes
v[i] accessO(1)True random access due to array layout
push_back (append)Amortized O(1)Reallocates occasionally
Insertion (front/mid)O(n)Requires shifting elements
ResizeO(n)Copy and reallocate

โš ๏ธ When capacity is exhausted, std::vector reallocates, usually doubling its capacity and copying all elements โ€” which invalidates pointers and iterators.

๐Ÿ’ป Example: Using std::vector for Rolling Average on Price Time Series

#include <vector>
#include <numeric>  // for std::accumulate

int main() {
    // Simulated price time series (e.g. 1 price per second)
    std::vector<double> prices;

    // Simulate appending tick data
    for (int i = 0; i < 10'000'000; ++i) {
        prices.push_back(100.0 + 0.01 * (i % 100)); // Fake tick data
    }

    // Compute rolling average over the last 5 prices
    size_t window_size = 5;
    std::vector<double> rolling_avg;

    for (size_t i = window_size; i < prices.size(); ++i) {
        double sum = std::accumulate(prices.begin() + i - window_size, prices.begin() + i, 0.0);
        rolling_avg.push_back(sum / window_size);
    }

    // Print the last rolling average
    std::cout << "Last rolling avg: " << rolling_avg.back() << std::endl;

    return 0;
}

๐Ÿง  Key Points:

  • std::vector<double> stores prices in contiguous heap memory, enabling fast iteration.
  • push_back appends elements efficiently until capacity needs to grow.
  • Access via prices[i] or iterators is cache-friendly.
  • This method is fast but not optimal for rolling computations โ€” a future section will show how std::deque or a circular buffer improves performance here.

2. std::deque: Double-Ended Queue

std::deque stands for “double-ended queue” โ€” a general-purpose container that supports efficient insertion and deletion at both ends. Unlike std::vector, which shines in random access, std::deque is optimized for dynamic FIFO (first-in, first-out) or sliding window patterns.

๐Ÿง  Stack vs Heap: Where Does Data Live?

  • Like std::vector, the std::deque object (its metadata) lives on the stack.
  • The underlying elements are allocated on the heap โ€” but not contiguously.

๐Ÿ’ก Consequences:

  • You can append or prepend large amounts of data without stack overflow.
  • Compared to std::vector, memory access is less cache-efficient due to fragmentation.

๐Ÿ“ฆ Contiguous Memory? No.

Unlike std::vector, which stores all elements in one continuous memory block, std::deque allocates multiple fixed-size blocks (typically 4KBโ€“8KB) and maintains a map (array of pointers) to those blocks.

Why this matters:

  • Appending to front/back is fast, since it just adds blocks.
  • But access patterns may be less cache-friendly, especially in tight loops.

โฑ Time Complexity

OperationTime ComplexityNotes
Random access d[i]O(1) (but slower than vector)Not contiguous, so slightly less efficient
push_back, push_frontO(1) amortizedEfficient at both ends
Insert in middleO(n)Requires shifting/moving like vector
ResizeO(n)Copying + new allocations

โš ๏ธ Internally more complex: uses a segmented memory model (think: mini-arrays linked together via pointer table).

โš™๏ธ Whatโ€™s Under the Hood?

A simplified mental model:

cppCopierModifierT** block_map;   // array of pointers to blocks
T* blocks[N];    // actual blocks on the heap
  • Each block holds a fixed number of elements.
  • deque manages which block and offset your index points to.

โœ… Strengths

  • Fast insertion/removal at both ends (O(1)).
  • No expensive shifting like vector.
  • Stable references to elements even after growth at ends (unlike vector).

โš ๏ธ Trade-offs

  • Not contiguous โ†’ worse cache locality than vector.
  • Random access is constant time, but slightly slower than vector[i].
  • Slightly higher memory overhead due to internal structure (block map + block padding).

๐Ÿ“ˆ Quant Finance Use Cases

  • Rolling windows on tick or quote streams (e.g. sliding 1-minute buffer).
  • Time-based buffers for short-term volatility, VWAP, or moving averages.
  • Event-driven systems where data is both consumed and appended continuously.

Example:

std::deque<double> price_buffer;
// push_front for new tick, pop_back to discard old

๐Ÿ’ฌ Typical Interview Questions

  1. “How would you implement a rolling average over a tick stream?”
    • A good answer involves std::deque for its O(1) append/remove at both ends.
  2. “Why not use std::vector if you’re appending at the front?”
    • Tests understanding of std::vector‘s O(n) front-insertion cost vs deque‘s O(1).
  3. “Explain the memory layout of deque vs vector, and how it affects performance.”
    • Looks for awareness of cache implications.

๐Ÿ’ป Code Example: Rolling Average with std::deque

cppCopierModifier#include <iostream>
#include <deque>
#include <numeric>

int main() {
    std::deque<double> price_window;
    size_t window_size = 5;

    for (int i = 0; i < 10; ++i) {
        double new_price = 100.0 + i;
        price_window.push_back(new_price);

        // Keep deque size fixed
        if (price_window.size() > window_size) {
            price_window.pop_front();  // Efficient O(1)
        }

        if (price_window.size() == window_size) {
            double avg = std::accumulate(price_window.begin(), price_window.end(), 0.0) / window_size;
            std::cout << "Rolling average: " << avg << std::endl;
        }
    }

    return 0;
}





๐Ÿง  Summary: deque vs vector





Featurestd::vectorstd::deque
Memory LayoutContiguous heapSegmented blocks on heap
Random AccessO(1), fastestO(1), slightly slower
Front Insert/RemoveO(n)O(1)
Back Insert/RemoveO(1) amortizedO(1)
Cache EfficiencyHighMedium
Quant Use CaseTime series, simulationsRolling buffers, queues





3. std::array: Fixed-Size Array





std::array is a stack-allocated, fixed-size container introduced in C++11. It wraps a C-style array with a safer, STL-compatible interface while retaining zero overhead. In quant finance, itโ€™s ideal when you know the size at compile time and want maximum performance and predictability.





๐Ÿง  Stack vs Heap: Where Does Data Live?





    

  • std::array stores all of its data on the stack โ€” no dynamic memory allocation.
    • 




  • The container itself is the data, unlike std::vector or std::deque, which hold pointers to heap-allocated buffers.
    • 






๐Ÿ’ก Consequences:





    

  • Blazing fast allocation/deallocation โ€” no malloc, no heap overhead.
    • 




  • Extremely cache-friendly โ€” everything is packed in a single memory block.
    • 




  • But: limited size (you canโ€™t store millions of elements safely).
    • 






๐Ÿ“ฆ Contiguous Memory? Yes.





std::array maintains a true contiguous layout, just like a C array. This enables:





    

  • Fast random access via arr[i]
    • 




  • Compatibility with C APIs
    • 




  • SIMD/vectorization-friendly memory layout
    • 






โฑ Time Complexity





OperationTime ComplexityNotes
Access a[i]O(1)Compile-time indexed if constexpr
SizeO(1)Always constant
InsertionN/AFixed size; no resizing





โš™๏ธ Whatโ€™s Under the Hood?





Conceptually:





cppCopierModifiertemplate<typename T, std::size_t N>
struct array {
    T elems[N];  // Inline data โ€” lives on the stack
};






    

  • No dynamic allocation.
    • 




  • No capacity concept โ€” just a fixed-size C array with better syntax and STL methods.
    • 






โœ… Strengths





    

  • Zero runtime overhead โ€” allocation is just moving the stack pointer.
    • 




  • Extremely fast access and iteration.
    • 




  • Safer than raw arrays โ€” bounds-safe with .at() and STL integration (.begin(), .end(), etc.).
    • 






โš ๏ธ Trade-offs





    

  • Fixed size โ€” canโ€™t grow or shrink at runtime.
    • 




  • Risk of stack overflow if size is too large.
    • 




  • Pass-by-value is expensive (copies all elements) unless using references.
    • 






๐Ÿ“ˆ Quant Finance Use Cases





    

  • Grids for finite difference/PDE solvers: std::array<double, N>
    • 




  • Precomputed factors (e.g., Greeks at fixed deltas)
    • 




  • Fixed-size historical feature vectors for ML signals
    • 




  • Small matrices or stat buffers in embedded systems or hot loops
    • 






// Greeks for delta hedging
std::array<double, 5> greeks = { 0.45, 0.01, -0.02, 0.005, 0.0001 };





๐Ÿ’ฌ Typical Interview Questions





    

  1. “Whatโ€™s the difference between std::array, std::vector, and C-style arrays?”

      

    • Tests understanding of memory layout, allocation, and safety.
      • 

    

    2. 




  2. “When would you prefer std::array over std::vector?”

      

    • Looking for โ€œwhen size is known at compile time and performance matters.โ€
      • 

    

    3. 




  3. “Whatโ€™s the risk of using large std::arrays?”

      

    • Looking for awareness of stack overflow and memory constraints.
      • 

    

    4. 






๐Ÿ’ป Code Example: Option Grid for PDE Solver





int main() {
    constexpr std::size_t N = 5; // e.g. 5 grid points in a discretized asset space
    std::array<double, N> option_values = {100.0, 101.5, 102.8, 104.2, 105.0};

    // Simple finite-difference operation: delta = (V[i+1] - V[i]) / dx
    double dx = 1.0;
    for (std::size_t i = 0; i < N - 1; ++i) {
        double delta = (option_values[i + 1] - option_values[i]) / dx;
        std::cout << "delta[" << i << "] = " << delta << std::endl;
    }

    return 0;
}










๐Ÿง  Summary: std::array vs vector/deque





Featurestd::arraystd::vectorstd::deque
MemoryStackHeap (contiguous)Heap (segmented)
SizeFixed at compile timeDynamic at runtimeDynamic at runtime
Allocation speedInstantSlow (heap)Slower (complex alloc)
Use casePDE grids, factorsTime series, simulationsRolling buffers
PerformanceMaxHighMedium










4. std::list / std::forward_list: Linked Lists





std::list and std::forward_list are the C++ STLโ€™s doubly- and singly-linked list implementations. While rarely used in performance-critical quant finance applications, they are still part of the container toolkit and may shine in niche situations involving frequent insertions/removals in the middle of large datasets โ€” but not for numeric data or tight loops.





๐Ÿง  Stack vs Heap: Where Does Data Live?





    

  • Both containers are heap-based: each element is dynamically allocated as a separate node.
    • 




  • The container object itself (head pointer, size metadata) is on the stack.
    • 




  • But the actual elements โ€” and their pointers โ€” are spread across the heap.
    • 






๐Ÿ’ก Consequences:





    

  • Extremely poor cache locality.
    • 




  • Each insertion allocates heap memory โ†’ slow.
    • 




  • Ideal for predictable insertion/removal patterns, not for numerical processing.
    • 






๐Ÿ“ฆ Contiguous Memory? Definitely Not.





Each element is a node containing:





    

  • The value
    • 




  • One (std::forward_list) or two (std::list) pointers
    • 






So memory looks like this:





[Node1] โ†’ [Node2] โ†’ [Node3] ...   (or doubly linked)





๐Ÿ’ฃ You lose any benefit of prefetching or SIMD. Iteration causes pointer chasing, leading to cache misses and latency spikes.





โฑ Time Complexity





OperationTime ComplexityNotes
Insert at frontO(1)Very fast, no shifting
Insert/remove in middleO(1) with iteratorNo shifting needed
Random access l[i]O(n)No indexing support
IterationO(n), but slowPoor cache usage





โš™๏ธ Whatโ€™s Under the Hood?





Each node contains:





    

  • value
    • 




  • next pointer (and prev for std::list)
    • 






So conceptually:





Node {
    T value;
    Node* next; // and prev, if doubly linked
};





    

  • std::list uses two pointers per node.
    • 




  • std::forward_list is lighter โ€” just one pointer โ€” but canโ€™t iterate backward.
    • 






โœ… Strengths





    

  • Efficient insertions/removals anywhere โ€” no element shifting like vector.
    • 




  • Iterator-stable: elements don’t move in memory.
    • 




  • Predictable performance for queue-like workloads with frequent mid-stream changes.
    • 






โš ๏ธ Trade-offs





    

  • Terrible cache locality โ€” each node is scattered.
    • 




  • No random access โ€” everything is via iteration.
    • 




  • Heavy memory overhead: ~2ร—โ€“3ร— the size of your data.
    • 




  • Heap allocation per node = slow.
    • 






๐Ÿ“ˆ Quant Finance Use Cases (Rare/Niche)





    

  • Managing a priority-sorted list of limit orders (insert/delete in mid-stream).
    • 




  • Event systems where exact insertion position matters.
    • 




  • Custom allocators or object pools in performance-tuned engines.
    • 






Usually, in real systems, this is replaced by:





    

  • std::deque for queues
    • 




  • std::vector for indexed data
    • 




  • Custom intrusive lists if performance really matters
    • 






๐Ÿ’ฌ Typical Interview Questions





    

  1. “Whatโ€™s the trade-off between std::vector and std::list?”

      

    • Looking for cache locality vs insert/delete performance.
      • 

    

    2. 




  2. “How would you store a stream of operations that require arbitrary insert/delete?”

      

    • Tests knowledge of when to use linked lists despite their cost.
      • 

    

    3. 




  3. “Why is std::list usually a bad idea for numeric workloads?”

      

    • Looking for insight into heap fragmentation and CPU cache behavior.
      • 

    

    4. 






๐Ÿ’ป Code Example: Linked List of Events





#include <iostream>
#include <list>

struct TradeEvent {
    double price;
    double quantity;
};

int main() {
    std::list<TradeEvent> event_log;

    // Simulate inserting trades
    event_log.push_back({100.0, 500});
    event_log.push_back({101.2, 200});
    event_log.push_front({99.5, 300});  // Insert at front

    // Iterate through events
    for (const auto& e : event_log) {
        std::cout << "Price: " << e.price << ", Qty: " << e.quantity << std::endl;
    }

    return 0;
}











๐Ÿง  Summary: std::list / std::forward_list





Featurestd::liststd::forward_list
MemoryHeap, scatteredHeap, scattered
AccessNo indexing, O(n)Forward only
Inserts/removalsO(1) with iteratorO(1), front-only efficient
Memory overheadHigh (2 pointers/node)Lower (1 pointer/node)
Cache efficiencyPoorPoor
Quant use caseRare (custom order books)Lightweight event chains





Unless you need fast structural changes mid-stream, these containers are rarely used in quant systems. But it’s important to know when and why they exist.

									            

        


		            

				                    

						                            
												                            0 comments
												                    

								            

		    








	
		
			                

										                        
							                                Bonds Pricer                        
					                

			
			
    

		            
Bonds

		
        
A Bonds Pricer Implementation in C++ with Quantlib

						

			by 										Clement D.
									
					

					    


    

        

            

									

Bond pricing is a fundamental skill for any quant. Whether you’re managing fixed-income portfolios or calculating risk metrics, understanding how a bond is priced is essential. In this article, weโ€™ll walk through the core formula, explore how clean and dirty prices differ, and implement it in C++ using QuantLib. How to build a bonds pricer?





1.The Bond Pricing Formula





When you buy a bond, youโ€™re essentially buying a stream of future cash flows: fixed coupon payments and the return of the face value at maturity. The price you pay for that stream depends on how much those future payments are worth today โ€” in other words, their present value.










๐Ÿ”น The Dirty Price: What Youโ€™re Really Paying





The dirty price (or full price) is the total value of the discounted future cash flows:





[math] \Large{\text{Dirty Price} = \sum_{i=1}^{N} \frac{C_i}{(1 + y)^{t_i}}} [/math]





Where:





    

  • [math]C_i[/math] is the cash flow at time [math]t_i[/math] โ€” typically the coupon, and the last cash flow includes the face value.
    • 




  • [math]y[/math] is the periodic yield, i.e., the annual yield divided by the number of coupon payments per year.
    • 




  • [math]t_i[/math] is the year fraction between today and the [math]i^\text{th}[/math] payment, calculated using day count conventions (like Actual/360, 30/360, etc.).
    • 






In simpler terms: money in the future is worth less than money today, and the discounting reflects that.










๐Ÿ”น Accrued Interest: What You Owe the Seller





Most bonds trade between coupon dates, meaning the seller has already โ€œearnedโ€ part of the next coupon. To make things fair, the buyer compensates them via accrued interest:





[math] \Large \text{Accrued Interest} = \text{Coupon} \times \frac{\text{Days since last payment}}{\text{Days in period}} [/math]





So if you buy the bond halfway through the coupon cycle, youโ€™ll owe the seller half the coupon. This ensures the next payment (which youโ€™ll receive in full) is properly split based on ownership time.










๐Ÿ”น Clean Price: The Market Quote





Bond prices are typically quoted clean, without accrued interest:





[math] \Large \text{Clean Price} = \text{Dirty Price} – \text{Accrued Interest} [/math]





This keeps things tidy when quoting and trading, and lets the system calculate accrued interest automatically behind the scenes.





Together, these three equations form the backbone of bond pricing. In the next section, weโ€™ll show how QuantLib brings them to life โ€” no need to write your own discounting engine.





3. Setting Up the Bond in QuantLib





Now that weโ€™ve covered the theory behind bond pricing, letโ€™s see how to implement it using QuantLib. One of QuantLibโ€™s biggest strengths is how it mirrors the real-world structure of financial instruments โ€” every component reflects a real aspect of how bonds are modeled, priced, and managed in production systems.





Hereโ€™s what we need to set up a bond in QuantLib:





๐Ÿ“… Calendar, Date, and Schedule





    

  • Calendar: Tells QuantLib which days are business days. This is crucial for calculating coupon dates and settlement dates correctly. Weโ€™ll use TARGET(), a commonly used calendar for euro-denominated instruments.
    • 




  • Date: QuantLibโ€™s custom date class used to define evaluation, issue, maturity, and coupon dates.
    • 




  • Schedule: Automatically generates all coupon dates between the start and maturity dates. You specify the frequency (e.g., semiannual), the calendar, and how to adjust dates if they fall on weekends or holidays.
    • 






In other words: this trio defines when things happen.





๐Ÿ’ต FixedRateBond





    

  • This is the actual bond object.
    • 




  • It takes in:

      

    • Settlement days (e.g., T+2),
      • 




    • Face value (e.g., 1000),
      • 




    • The coupon schedule,
      • 




    • The fixed coupon rate(s),
      • 




    • Day count convention (e.g., Actual/360),
      • 




    • Date adjustment rules.
      • 

    

    • 




  • Once constructed, it contains all the future cash flows, knows when they occur, and how much they are.
    • 






Think of FixedRateBond as your contractual definition of the bond.





๐Ÿ“ˆ YieldTermStructure





    

  • This represents the discount curve.
    • 




  • You can define it using:

      

    • A flat yield (e.g., 4.5% across all maturities),
      • 




    • A bootstrap from market instruments (swaps, deposits, etc.),
      • 




    • Or even a custom curve from CSV or historical data.
      • 

    

    • 




  • QuantLib uses this curve to discount each cash flow to present value.
    • 






This is your interest rate environment โ€” essential for pricing.





๐Ÿง  DiscountingBondEngine





    

  • This is the pricing engine: the part of QuantLib that ties it all together.
    • 




  • Once you set the bondโ€™s pricing engine to a DiscountingBondEngine, it knows how to price it using the curve.
    • 




  • It computes:

      

    • The dirty price,
      • 




    • The clean price,
      • 




    • The accrued interest,
      • 




    • And other risk measures like duration and convexity.
      • 

    

    • 






You can think of it as the calculator that applies all the math we just discussed.





4. Implementation





Let’s implement:





#include <ql/quantlib.hpp>
#include <iostream>
#include <iomanip>

using namespace QuantLib;

int main() {
    // Set the evaluation date
    Date today(24, June, 2025);
    Settings::instance().evaluationDate() = today;

    // Bond parameters
    Real faceValue = 1000.0;
    Rate couponRate = 0.05; // 5%
    Date maturity(24, June, 2030);
    Frequency frequency = Semiannual;
    Integer settlementDays = 2;
    DayCounter dayCounter = Actual360();
    BusinessDayConvention convention = Unadjusted;
    Calendar calendar = TARGET();

    // Build the schedule of coupon payments
    Schedule schedule(today, maturity, Period(frequency), calendar,
                      convention, convention, DateGeneration::Backward, false);

    // Create the fixed-rate bond
    FixedRateBond bond(settlementDays, faceValue, schedule,
                       std::vector<Rate>{couponRate}, dayCounter, convention);

    // Build a flat yield curve (4.5%)
    Handle<YieldTermStructure> yieldCurve(
        boost::make_shared<FlatForward>(today, 0.045, dayCounter));

    // Set the pricing engine using the discounting curve
    bond.setPricingEngine(boost::make_shared<DiscountingBondEngine>(yieldCurve));

    // Output the results
    std::cout << std::fixed << std::setprecision(2);
    std::cout << "Clean Price   : " << bond.cleanPrice() << std::endl;
    std::cout << "Dirty Price   : " << bond.dirtyPrice() << std::endl;
    std::cout << "Accrued Interest: " << bond.accruedAmount() << std::endl;

    return 0;
}





    

  1. 






After compilation and run, we get:





โžœ  build ./pricer  
Clean Price   : 102.01
Dirty Price   : 102.04
Accrued Interest: 0.03










๐Ÿ”น Clean Price = 102.01





This is the market-quoted price of the bond, excluding any interest that has accrued since the last coupon date.
It means that the bond is trading at 102.01% of its face value โ€” i.e., ยฃ1,020.10 for a bond with a ยฃ1,000 face value.










๐Ÿ”น Dirty Price = 102.04





This is the actual amount you’d pay if you bought the bond today.
It includes both:





    

  • The clean price (102.01), and
    • 




  • The interest the seller has “earned” since the last coupon.
    • 






So:




[math]

\text{Dirty Price} = \text{Clean Price} + \text{Accrued Interest} = 102.01 + 0.03 = 102.04

[/math]










๐Ÿ”น Accrued Interest = 0.03





This is the amount of coupon interest that has accrued since the last coupon date but hasnโ€™t been paid yet.
The buyer pays this to the seller because the seller has held the bond for part of the current coupon period.





In your case, itโ€™s 0.03% of face value, or ยฃ0.30 on a ยฃ1,000 bond โ€” meaning youโ€™re very close to the previous coupon date.

									            

        


		            

				                    

						                            
												                            0 comments
												                    

								            

		    








	
		
			                

										                        
							                                Value at Risk                        
					                

			
			
    

		            
VaR

		
        
Value at Risk (VaR): Definition, Equation, and C++ Implementation

						

			by 										Clement D.
									
					

					    


    

        

            

									

Value at Risk (VaR) is a statistical measure used to quantify the level of financial risk within a portfolio over a specific time frame. It answers the question: โ€œWhat is the maximum expected loss with a given confidence level over a given time horizon?โ€





Widely used in trading, risk management, and regulatory frameworks (e.g., Basel III), VaR helps institutions understand the tail risk of their holdings.





1.Formal Definition and Example





Value at Risk (VaR) at confidence level [math] \alpha [/math] (e.g., 95% or 99%) is defined as the threshold loss value L such that:





[math]\large \mathbb{P}(\text{Loss} > L) = 1 – \alpha[/math]





Or, written differently:





[math]\large \text{VaR}_\alpha = \inf \left\{ l \in \mathbb{R} : \mathbb{P}(\text{Loss} \leq l) \geq \alpha \right\}[/math]





Some examples to make it a bit more concrete:






๐Ÿ”น Example 1 โ€” Conservative Institution







For a 1-day 99% VaR of $5 million, you expect that on 99% of trading days, your portfolio will not lose more than $5 million.
However, in the remaining 1% of days (about 2-3 days per year), the losses could exceed $5 million.












๐Ÿ”น Example 2 โ€” Moderate Risk Portfolio







For a 10-day 95% VaR of โ‚ฌ2 million, there is a 95% chance that over any 10-day period, losses will not exceed โ‚ฌ2 million.
Conversely, there’s a 5% chance (about one 10-day period in 20) that you could lose more than โ‚ฌ2 million.












๐Ÿ”น Example 3 โ€” Intraday Trading Desk







For a 1-hour 90% VaR of ยฃ100,000, you are 90% confident that in any given hour, the desk will not lose more than ยฃ100,000.
But in 1 out of every 10 hours, losses could exceed ยฃ100,000.












2. How Traders Like to Vizualize VaR?





traders and risk managers often use VaR summary tables to visualize potential losses across different confidence levels, time horizons, and portfolio segments. Here’s how it’s typically structured:










โœ…  Standard VaR Table





Confidence LevelTime HorizonVaR (USD)
95%1 Day$1,200,000
99%1 Day$2,300,000
95%10 Days$3,800,000
99%10 Days$7,200,000





    

  • This format lets traders quickly gauge the magnitude of risk at different percentiles.
    • 




  • The 10-day VaR is often used for regulatory reporting (e.g., Basel rules).
    • 




  • The square-root-of-time rule is sometimes used to scale 1-day VaR to longer horizons.
    • 











โœ… Portfolio Breakdown Table





Desk / Asset Class1-Day 95% VaR1-Day 99% VaR
Equities$500,000$850,000
Fixed Income$300,000$620,000
FX$250,000$470,000
Commodities$150,000$280,000
Total$980,000$1,750,000





    

  • Often used in daily risk reports.
    • 




  • May include a correlation adjustment between asset classes (not a simple sum).
    • 











โœ… 3. VaR vs. PnL Table (Backtesting)





DateActual PnL1-Day 99% VaRBreach?
2025-06-18-$3.5M$2.8Mโœ…
2025-06-19-$1.2M$2.8MโŒ
2025-06-20-$2.9M$2.8Mโœ…





    

  • Used for Value at Risk backtesting, checking how often real losses exceed VaR.
    • 




  • Frequent breaches might indicate model underestimation of tail risk.
    • 






3. The Different Ways to Calculate VaR





To calculate Value at Risk (VaR), different methods are used depending on the assumptions you’re willing to make and the data available. Some approaches rely on statistical models, while others lean purely on historical data:





–The Parametric (Variance-Covariance) VaR
Assumes returns are normally distributed and calculates VaR using the portfolioโ€™s mean, standard deviation, and the Z-score of the desired confidence level.





-The Historical Simulation VaR
Uses actual historical returns to simulate potential losses, without assuming any specific return distribution.





-The Monte Carlo Simulation VaR
Generates thousands of possible return scenarios using stochastic models (e.g., Geometric Brownian Motion) to estimate potential losses under a wide range of outcomes.





4. A Zoom on Parametric VaR





The Parametric VaR method assumes that portfolio returns are normally distributed, making it one of the fastest and most widely used VaR models in practice.





The general formula is:





[math]\large \text{VaR}_\alpha = \mu + z_\alpha \cdot \sigma[/math]





Where:





    

  • [math]\mu[/math] = Expected return (often set to 0 for short-term horizons)
    • 




  • [math]\sigma[/math] = Standard deviation (volatility) of portfolio returns
    • 




  • [math]z_\alpha[/math] = Z-score for the chosen confidence level (e.g., -1.6449 for 95%, -2.3263 for 99%)
    • 






This formula gives you the threshold loss you should not exceed with probability [math]\alpha[/math].
For example, at 99% confidence, Parametric VaR tells you: โ€œThere is only a 1% chance that the portfolio will lose more than VaR in a day.โ€





Because it’s simple and fast to compute, Parametric VaR is used in real-time risk monitoring, but it can underestimate tail risk when returns deviate from normality (e.g., skew or fat tails).





5. Implement Parametric VaR in C++





Here is a basic implementation, like we did elsewhere for greeks, without using any libraries:





#include <iostream>
#include <vector>
#include <cmath>
#include <numeric>
#include <stdexcept>

// Compute the mean of returns
double computeMean(const std::vector<double>& returns) {
    double sum = std::accumulate(returns.begin(), returns.end(), 0.0);
    return sum / returns.size();
}

// Compute standard deviation of returns
double computeStdDev(const std::vector<double>& returns, double mean) {
    double accum = 0.0;
    for (double r : returns) {
        accum += (r - mean) * (r - mean);
    }
    return std::sqrt(accum / (returns.size() - 1));
}

// Get Z-score for given confidence level
double getZScore(double confidenceLevel) {
    if (confidenceLevel == 0.95) return -1.64485;
    if (confidenceLevel == 0.99) return -2.32635;
    throw std::invalid_argument("Unsupported confidence level");
}

// Compute Parametric VaR
double computeParametricVaR(const std::vector<double>& returns,
                            double confidenceLevel,
                            double portfolioValue) {
    double mean = computeMean(returns);  // often assumed 0 for short-term VaR
    double stdDev = computeStdDev(returns, mean);
    double z = getZScore(confidenceLevel);

    double var = portfolioValue * (mean + z * stdDev);
    return std::abs(var); // Loss is a positive number
}

int main() {
    std::vector<double> returns = {-0.01, 0.003, 0.0045, -0.002, 0.005}; // Daily returns
    double confidenceLevel = 0.99;
    double portfolioValue = 1'000'000; // $1M

    double var = computeParametricVaR(returns, confidenceLevel, portfolioValue);
    std::cout << "1-day 99% Parametric VaR: $" << var << std::endl;

    return 0;
}





Let’s compile it:





โžœ  build cmake ..
-- Configuring done (0.1s)
-- Generating done (0.0s)
-- Build files have been written to: /Users/clementdaubrenet/var/build
โžœ  build make       
[ 50%] Building CXX object CMakeFiles/var.dir/var.cpp.o
[100%] Linking CXX executable var
[100%] Built target var





And run it:





โžœ  build ./var           
1-day 99% Parametric VaR: $14530.1





The 1-day 99% VaR for this portfolio is $14530.1.

									            

        


		            

				                    

						                            
												                            0 comments
												                    

								            

		    








	
		
			                

										                        
							                                Delta Greek Calculation                        
					                

			
			
    

		            
Greeks

		
        
Option Greeks in C++: Delta Calculation Explained

						

			by 										Clement D.
									
					

					    


    

        

            

									

Delta is one of the most important Greeks in options pricing. It measures how sensitive an optionโ€™s price is to changes in the price of the underlying asset. How to perform a delta calculation? Let’s derive and it and calculate it in C++.





1. Derive the Delta Formula for a Call Option





Delta is defined as the rate of change of an optionโ€™s price with respect to changes in the underlying asset price. In mathematical terms, it’s the first partial derivative of the option value V with respect to the underlying asset price S:





[math] \Large{\Delta = \frac{\partial V}{\partial S}} [/math]





Under the Black-Scholes framework, the price of a European call option is given by:





[math]\Large{C = S N(d_1) – K e^{-rT} N(d_2)}[/math]





Where:





    

  • S is the current spot price of the underlying
    • 




  • K is the strike price
    • 




  • r is the risk-free interest rate
    • 




  • T is the time to maturity
    • 




  • N(โ‹…) is the cumulative distribution function of the standard normal distribution
    • 




  • [math]d_1 = \frac{\ln(S/K) + \left(r + \frac{\sigma^2}{2} \right) T}{\sigma \sqrt{T}}[/math]
    • 




  • [math]d_2 = d_1 – \sigma \sqrt{T}[/math]
    • 






To compute the Delta of a call option, we differentiate the Black-Scholes formula with respect to S:





[math]\Large \frac{\partial C}{\partial S} = \frac{\partial}{\partial S} \left[ S N(d_1) – K e^{-rT} N(d_2) \right][/math]





Only the first term depends on S directly. The derivative of the first term requires the chain rule, since d1โ€‹ is a function of S as well. Thus:





[math]\Large \frac{\partial C}{\partial S} = N(d_1)[/math]





Therefore, the Delta of a European call option is: 





[math]\Large \Delta_{\text{call}} = N(d_1)[/math]










2. Derive the Delta Formula For a Put Option





For a put option, we apply the same steps to the Black-Scholes formula for puts:





[math]\Large P = K e^{-rT} N(-d_2) – S N(-d_1)[/math]





Where:





    

  • S is the current spot price of the underlying
    • 




  • K is the strike price
    • 




  • r is the risk-free interest rate
    • 




  • T is the time to maturity
    • 




  • N(โ‹…) is the cumulative distribution function of the standard normal distribution
    • 




  • [math]d_1 = \frac{\ln(S/K) + \left(r + \frac{\sigma^2}{2} \right) T}{\sigma \sqrt{T}}[/math]
    • 




  • [math]d_2 = d_1 – \sigma \sqrt{T}[/math]
    • 






Differentiating:





[math]\Large \frac{\partial P}{\partial S} = -N(-d_1) = N(d_1) – 1[/math]





Hence, the Delta of a European put option is:





[math]\Large \Delta_{\text{put}} = N(d_1) – 1[/math]





3. Delta Calculation in C++ Without Library





We can simply implement the recipes above to calculate delta call and delta put:





#include <cmath>
#include <iostream>

// Cumulative normal distribution function
double norm_cdf(double x) {
    return 0.5 * std::erfc(-x / std::sqrt(2));
}

double black_scholes_delta(
    bool is_call,
    double S,     // Spot
    double K,     // Strike
    double T,     // Time to maturity
    double r,     // Risk-free rate
    double sigma  // Volatility
) {
    double d1 = (std::log(S / K) + (r + 0.5 * sigma * sigma) * T)
                / (sigma * std::sqrt(T));

    if (is_call) {
        return norm_cdf(d1);
    } else {
        return norm_cdf(d1) - 1.0;
    }
}






Then:





int main() {
    double S = 100.0;
    double K = 100.0;
    double T = 1.0;
    double r = 0.05;
    double sigma = 0.2;

    double delta_call = black_scholes_delta(true, S, K, T, r, sigma);
    double delta_put  = black_scholes_delta(false, S, K, T, r, sigma);

    std::cout << "Call Delta: " << delta_call << std::endl;
    std::cout << "Put Delta: "  << delta_put  << std::endl;
    return 0;
}










This is implements the analytical formula for Delta explicitly. It’s lightweight and very fast, great for educational or performance-critical code. It assumes flat rates and no dividends.





But what if you want something less generic?





4. Delta Calculation in C++ With Quantlib





Well, we can use the bazooka:





#include <ql/quantlib.hpp>
#include <iostream>

using namespace QuantLib;

int main() {
    Calendar calendar = TARGET();
    Date today(23, June, 2025);
    Settings::instance().evaluationDate() = today;

    // Common option parameters
    Real underlying = 100.0;
    Real strike = 100.0;
    Spread dividendYield = 0.0;
    Rate riskFreeRate = 0.05;
    Volatility volatility = 0.20;
    Date maturity = today + Period(1, Years);
    DayCounter dayCounter = Actual365Fixed();

    // Market data handles
    Handle<Quote> underlyingH(boost::shared_ptr<Quote>(new SimpleQuote(underlying)));
    Handle<YieldTermStructure> flatDividendTS(boost::shared_ptr<YieldTermStructure>(
        new FlatForward(today, dividendYield, dayCounter)));
    Handle<YieldTermStructure> flatRiskFreeTS(boost::shared_ptr<YieldTermStructure>(
        new FlatForward(today, riskFreeRate, dayCounter)));
    Handle<BlackVolTermStructure> flatVolTS(boost::shared_ptr<BlackVolTermStructure>(
        new BlackConstantVol(today, calendar, volatility, dayCounter)));

    boost::shared_ptr<BlackScholesMertonProcess> bsmProcess(
        new BlackScholesMertonProcess(underlyingH, flatDividendTS, flatRiskFreeTS, flatVolTS));

    // Loop over Call and Put options
    for (auto type : {Option::Call, Option::Put}) {
        std::string typeStr = (type == Option::Call ? "Call" : "Put");

        boost::shared_ptr<StrikedTypePayoff> payoff(new PlainVanillaPayoff(type, strike));
        boost::shared_ptr<Exercise> exercise(new EuropeanExercise(maturity));
        VanillaOption option(payoff, exercise);

        option.setPricingEngine(boost::shared_ptr<PricingEngine>(
            new AnalyticEuropeanEngine(bsmProcess)));

        std::cout << typeStr << " Delta: " << option.delta() << std::endl;
    }

    return 0;
}










โœ… The QuantLib implementation:





    

  • Wraps the same math, but within a complete pricing engine (AnalyticEuropeanEngine).
    • 




  • Uses market objects like Quote, YieldTermStructure, and BlackVolTermStructure, which can evolve over time.
    • 




  • Supports dividends, yield curves, non-constant volatility, and more.
    • 






In a word, it’s heavier but more flexible and extensible, making it better suited for production systems or derivatives desks.





5. Let’s Run Both and Compare?





Let’s start with the first version, after compiling and running:





โžœ  ./delta_vanilla                                                              
Call Delta: 0.636831
Put Delta: -0.363169





Now, the Quantlib version:





โžœ  build ./delta_quantlib 
Call Delta: 0.636831
Put Delta: -0.363169





The exact same!





What about execution time?





For the first version, it’s instant, <<0.01s:





โžœ  build time ./delta_vanilla 

./delta_vanilla  0.00s user 0.00s





For the Quantlib version, ~0.01s, slightly slower:





โžœ  build time ./delta_quantlib

./delta_quantlib  0.00s user 0.01s





4. Summarize Both Implementations for Delta Calculation










FeatureVanillaQuantLib
Analytical Deltaโœ… Yesโœ… Yes
Dividend supportโŒ Noโœ… Yes
Yield curve supportโŒ Noโœ… Yes
Vol surface supportโŒ Noโœ… Yes
Performanceโšก Very fast๐Ÿข Slower (but flexible)
Good for learningโœ… Absolutelyโœ… But more complex
Good for real-world desksโŒ Not sufficientโœ… Industry standard





If you’re writing an educational article or prototyping, the vanilla version is perfect. If you’re integrating with a full market data and pricing system, QuantLib is the go-to.





Code of the article with Readme for compilation and execution:





https://github.com/cppforquants/delta/tree/main






									            

        


		            

				                    

						                            
												                            0 comments
												                    

								            

		    








	
		
			                

										                        
							                                tree shap model                        
					                

			
			
    

		            
Credit Risk

		
        
Attribution Modeling for Credit PnL Using TreeSHAP

						

			by 										Clement D.
									
					

					    


    

        

            

									

In modern credit trading, understanding why PnL moves is just as important as tracking that it moved.
When billions of data points flow through trading systems daily, intuition and heuristics fall short.
This article presents a scalable method for attributing PnL changes to specific credit risk drivers.
We model feature deltas โ€” changes in exposure, spreads, and counterparty metrics โ€” across time.






We train a LightGBM model to learn nonlinear relationships between deltas and ฮ”PnL.
Then, we apply TreeSHAP to extract the top features responsible for each credit-related PnL movement.
The result is a transparent, auditable, per-entity view of what caused the last PnL move.
This is not prediction, itโ€™s attribution โ€” a diagnostic engine for credit risk.





1. Introduction: The Need for Attribution, Not Prediction





Credit-related PnL refers to profit and loss driven by credit risk exposure โ€” such as changes in counterparty credit spreads, rating downgrades, or shifts in default probabilities.
It captures the financial impact of events like widening CDS, growing exposures, or collateral changes, rather than market or operational effects.






For traders, a sudden loss tied to a counterparty matters less than knowing which risk factor caused it.
For quants, understanding nonlinear relationships between inputs and PnL is key to model trust and validation.
And for risk teams, attribution supports regulatory reporting, stress testing, and early warnings.
But with data at the scale of billions of deltas, traditional methods fail to offer clear answers.
We propose a hybrid pipeline: use Python to train a LightGBM model on feature deltas and PnL deltas.
Then apply TreeSHAP to explain which deltas caused most of the PnL movement.
For real-time inference, we port the model and SHAP logic to a high-performance C++ backend.
The result: an explainable, scalable, and production-ready system for credit PnL attribution at scale.





2. Data Design: Tracking the Right Deltas





Every 15 minutes, a new snapshot arrives โ€” a high-dimensional vector (10,000+ features) representing the full state of the trading system: exposure, spreads, counterparty credit ratings, margin calls, instrument properties, and more.





This raw data isn’t yet meaningful. To extract signal, we compute deltas: changes in features between consecutive time windows. These represent what moved.





Alongside the feature deltas, we compute the PnL delta, giving us a supervised signal to explain: what caused the profit or loss change since the last window?





We filter and engineer meaningful features โ€” credit spreads, counterparty ratings, net exposure, sector-level aggregates โ€” to reduce noise and emphasize credit drivers.





Using deltas instead of raw levels is key. Most credit features are auto-correlated or slow-moving โ€” a high exposure value tells us less than a sudden increase in exposure.





The delta of a credit spread or notional tells us more about risk events, decisions, or systemic shifts than the absolute value alone.





By modeling dPnL ~ ฮ”_features, we move from static snapshots to causal attribution.





๐Ÿ“„ Mocked Extract of Delta Dataset





timestampentityฮ”_exposureฮ”_cds_bnpฮ”_rating_scoreฮ”_sector_riskฮ”PnL
2025-06-23 10:00:00trader_01+2.3M+0.018-1+0.07-4.21M
2025-06-23 10:15:00trader_01-1.1M-0.0050-0.03+2.73M
2025-06-23 10:30:00trader_01+0.4M+0.00200-1.00M







    

  • ฮ”_exposure: Notional delta to counterparties
    • 




  • ฮ”_cds_bnp: Change in BNP CDS (in decimal)
    • 




  • ฮ”_rating_score: Discrete rating movement (e.g., 1 = downgrade)
    • 




  • ฮ”_sector_risk: Change in weighted average sector risk score
    • 




  • dPnL: Total change in credit-related PnL
    • 








We will drastically increase the dimensionality of those with 10k columns and 1 million rows.





We generate millions of entries in “delta_features.csv” and millions of delta PnLs in “delta_pnl.csv”





3. Modeling: Training a Gradient Boosted Tree on Deltas





To prepare for TreeSHAP attribution in this credit delta โ†’ PnL delta setup, youโ€™ll want to use a LightGBM regression model that emphasizes explainability, stability, and performance.





โœ… Goal:





    

  • Train a LightGBM model (ฮ”features โ†’ ฮ”PnL)
    • 




  • Export the model (model.txt)
    • 




  • Optionally export samples for testing C++ inference
    • 






๐Ÿง  File: train_model.py










import lightgbm as lgb
import pandas as pd

# Load delta feature and delta PnL data
X = pd.read_csv("delta_features.csv")
y = pd.read_csv("delta_pnl.csv")

# Train model
params = {
    "objective": "regression",
    "metric": "rmse",
    "learning_rate": 0.03,
    "num_leaves": 64,
    "feature_fraction": 0.6,
    "bagging_fraction": 0.8,
    "bagging_freq": 5,
    "lambda_l1": 0.1,
    "lambda_l2": 1.0,
    "min_data_in_leaf": 50,
    "verbose": -1
}
model = lgb.train(lgb.Dataset(X, label=y), params, num_boost_round=200)

# Save model
model.save_model("model.txt")

# Export a row (e.g., for test inference)
row_index = 12345
X.iloc[[row_index]].to_csv("sample_row.csv", index=False)
y.iloc[[row_index]].to_csv("sample_label.csv", index=False)





Then we can run the inferrence:





import pandas as pd
import shap
import lightgbm as lgb

# Load model and data
model = lgb.Booster(model_file="model.txt")          # Trained LGBM model
X = pd.read_csv("sample_row.csv")         # Latest 15-min features (deltas)
y_pnl = pd.read_csv("sample_label.csv")          # Actual PnL deltas (optional)

# Run SHAP
explainer = shap.TreeExplainer(model)
shap_values = explainer.shap_values(X)

# For a specific row (e.g. last 15-min observation)
row = 0
shap_row = shap_values[row]

# Create attribution table
explained = pd.DataFrame({
    "feature": X.columns,
    "shap_value": shap_row,
    "feature_value": X.iloc[row]
}).sort_values(by="shap_value", key=abs, ascending=False)

# Add cumulative contribution
explained["cum_impact"] = explained["shap_value"].cumsum()
total_impact = abs(explained["shap_value"]).sum()
explained["cum_percent"] = abs(explained["shap_value"]).cumsum() / total_impact

# Output: top features responsible for 90% of the PnL move
top_contributors = explained[explained["cum_percent"] <= 0.9]
print(top_contributors)





4. Results





๐Ÿ“ฆ Output Example





      feature      shap_value  feature_value  cum_impact  cum_percent
23  spread_delta          0.68          12.0         0.68     0.42
14     curve_1y          0.51           3.2         1.19     0.72
9   exposure_bond         0.33          54.1         1.52     0.92





This shows:





    

  • The features most responsible for the ฮ”PnL for that 15-min slice
    • 




  • Their raw SHAP impact
    • 




  • How much of the total impact they cumulatively explain
    • 







									            

        


		            

				                    

						                            
												                            0 comments
												                    

								            

		    








	
		
			                

										                        
							                                Implied Volatility                        
					                

			
			
    

		            
Volatility

		
        
Compute the Implied Volatility for a Call Option in C++

						

			by 										Clement D.
									
					

					    


    

        

            

									

Volatility is a cornerstone concept in finance, used to measure the degree of variation in the price of a financial instrument. Two of the most commonly referenced types are Implied Volatility (IV) and Historical Volatility (HV), and while they both aim to assess risk, they do so from different perspectives.





Implied Volatility is forward-looking. Derived from option prices using models like Black-Scholes, it reflects the marketโ€™s expectations of future price movement. Higher IV typically signals greater anticipated risk or uncertainty.





HV is objective and purely data-driven, while IV incorporates market sentiment and supply-demand dynamics in the options market.





1. Derive Black-Scholes PDE Solution for European Call Options





To derive implied volatility from the Black-Scholes model, you’re essentially trying to find the volatility (ฯƒ) that, when plugged into the Black-Scholes formula, yields the market price of the option.





Let’s start from Black-Scholes.





[math] 
\Large{dS_t = \mu S_t \, dt + \sigma S_t \, dW_t}
[/math]





is a stochastic differential equation (SDE) โ€” it’s the core model behind how asset prices evolve over time in the Black-Scholes framework.










๐Ÿ” Term-by-Term Breakdown





SymbolMeaning
[math] S_tโ€‹ [/math]The price of the asset (e.g., stock) at time ttt
[math] dS_t [/math]The infinitesimal change in the asset price over a small time interval
[math]  \mu [/math] The drift โ€” average rate of return (expected % growth per unit time)
dtA very small time interval (infinitesimal)
[math] \sigma [/math]The volatility โ€” standard deviation of returns per unit time (how “noisy” the price is)
[math] dWt [/math]A random shock, also called a Wiener process or Brownian motion increment. Represents randomness.





The formula models price changes as having two components:





A deterministic part:





[math] \Large{\mu S_t \, dt} [/math]





This is the trend โ€” the predictable, average growth over time.





A stochastic part:





[math]  \Large{\sigma S_t \, dW_t} [/math]





This is the randomness โ€” the noise or uncertainty in price movements.





    

  • [math] \sigma [/math] scales the size of the randomness.
    • 




  • [math] dW_tโ€‹ [/math] brings in randomness from a standard Brownian motion: the mean is 0 and the variance is dt 
    • 






A first thing to do to reach the implied volatility question is to notice that to price options, we donโ€™t use [math] mu [/math] because investors care about risk-adjusted returns.





Instead, we use r for the risk-neutral measure [math] \mathbb{Q} [/math], where:





[math] \Large{dS_t = r S_t \, dt + \sigma S_t \, dW_t^{\mathbb{Q}}} [/math]





    

  • The drift becomes r, the risk-free rate.
    • 




  • The randomness remains via [math] \sigma [/math] and [math] dW_t [/math]
    • 






๐Ÿ” Let’s define the call option price






[math]\Large{ C(t,S_tโ€‹)=priceย ofย aย callย option} [/math]





After a bit of work that we will not do in this article, we can derive the Black-Scholes PDE:





[math] \Large{\frac{\partial C}{\partial t} + r S_t \frac{\partial C}{\partial S} + \frac{1}{2} \sigma^2 S_t^2 \frac{\partial^2 C}{\partial S^2} – r C = 0} [/math]





Solving this PDE with the terminal condition: 





[math] \Large{C(T, S_T) = \max(S_T – K, 0)} [/math]





gives us the option price at time t=0:





[math] \Large{C(\sigma) = S_0 \Phi(d_1) – K e^{-rT} \Phi(d_2)} [/math]





where: 





[math] \Large{d_1 = \frac{\ln(S_0 / K) + (r + \frac{1}{2} \sigma^2) T}{\sigma \sqrt{T}}} [/math]





And





[math] \Large{d_2 = d_1 – \sigma \sqrt{T}} [/math]





2. Determine the Equation to Solve to get the Implied Volatility 





We now take the market-observed option price [math] C_{market} [/math] and ask:







โ€œWhat value of  [math] \sigma [/math] makes the Black-Scholes price equal the market price?โ€







So we define the equation to solve:





[math] \Large{f(\sigma) = C(\sigma)  \, – C_{\text{market}} = 0} [/math]





Since C(ฯƒ)C(\sigma)C(ฯƒ) involves ฮฆ(d1)\Phi(d_1)ฮฆ(d1โ€‹) and d1d_1d1โ€‹ contains ฯƒ\sigmaฯƒ both in the numerator and denominator, there is no closed-form algebraic solution.





We solve it using numerical methods:





    

  • Brentโ€™s method (robust)
    • 




  • Newton-Raphson (faster, requires Vega)
    • 




  • Bisection (simple, slower)
    • 






What does it mean to solve that equation in a real-world context?





๐Ÿงฎ Example





Letโ€™s say the following is true today:





ParameterValue
[math] S_0 [/math]โ€‹ (stock price)$100
K (strike)$105
T (time to maturity)30 days โ‰ˆ 0.0822 years
r (risk-free rate)5% = 0.05
[math] C_{market} [/math] option price from market$2.50





๐Ÿ‘‡ Now We Need To Solve:





[math] \Large{f(\sigma) = C(\sigma) – 2.50 = 0} [/math]





You plug different values of [math] \sigma [/math] into the Black-Scholes formula: 





[math] \Large{C(\sigma) = S_0 \Phi(d_1) – K e^{-rT} \Phi(d_2)} [/math]





where:





[math] \Large{d_1 = \frac{\ln(S_0 / K) + (r + \frac{1}{2} \sigma^2) T}{\sigma \sqrt{T}}} [/math]





[math] \Large{d_2 = d_1 – \sigma \sqrt{T}} [/math]





And iterate until:





 [math] \Large{C(\sigma) \approx 2.50} [/math].





So, how do we iterate to solve the volatility equation?





3. Use Quantlib to Solve the Implied Volatility Equation 





Quantlib will solve that equation with a numerical method.

Here is the code for the example above:





#include <ql/quantlib.hpp>
#include <iostream>

using namespace QuantLib;

int main() {
    Calendar calendar = TARGET();
    Date today = Date::todaysDate();
    Settings::instance().evaluationDate() = today;

    // Option parameters
    Option::Type optionType = Option::Call;
    Real underlying = 100.0;
    Real strike = 105.0;
    Rate riskFreeRate = 0.05;
    Volatility initialVol = 0.20; // Just an initial guess
    Date maturity = calendar.advance(today, Period(30, Days));  // 30-day maturity
    Real marketPrice = 2.50;

    DayCounter dayCounter = Actual365Fixed();
    Handle<Quote> underlyingH(boost::shared_ptr<Quote>(new SimpleQuote(underlying)));

    // Constructing the yield curve and flat volatility surface
    Handle<YieldTermStructure> flatTermStructure(
        boost::shared_ptr<YieldTermStructure>(
            new FlatForward(today, riskFreeRate, dayCounter)));

    Handle<BlackVolTermStructure> flatVolTS(
        boost::shared_ptr<BlackVolTermStructure>(
            new BlackConstantVol(today, calendar, initialVol, dayCounter)));

    boost::shared_ptr<StrikedTypePayoff> payoff(
        new PlainVanillaPayoff(optionType, strike));

    boost::shared_ptr<Exercise> exercise(
        new EuropeanExercise(maturity));

    EuropeanOption option(payoff, exercise);

    Handle<YieldTermStructure> flatDividendTS(
    boost::shared_ptr<YieldTermStructure>(
        new FlatForward(today, 0.0, dayCounter)));

    boost::shared_ptr<BlackScholesMertonProcess> bsmProcess(
        new BlackScholesMertonProcess(underlyingH, 
                                      flatDividendTS, 
                                      flatTermStructure, 
                                      flatVolTS));

    // Calculate implied volatility
    try {
        Volatility impliedVol = option.impliedVolatility(
            marketPrice, bsmProcess, 1e-6, 100, 0.0001, 4.0);
        std::cout << "Implied Volatility: " << impliedVol << std::endl;
    } catch (std::exception& e) {
        std::cerr << "Error calculating implied volatility: " << e.what() << std::endl;
    }

    return 0;
}






after:





brew install quantlib





and setting up your CMakeLists.txt:





cmake_minimum_required(VERSION 3.10)
project(QuantLibImpliedVolExample)

set(CMAKE_CXX_STANDARD 17)

find_package(PkgConfig REQUIRED)
pkg_check_modules(QUANTLIB REQUIRED QuantLib)

include_directories(${QUANTLIB_INCLUDE_DIRS})
link_directories(${QUANTLIB_LIBRARY_DIRS})

add_executable(implied_vol src/volatility.cpp)
target_link_libraries(implied_vol ${QUANTLIB_LIBRARIES})





and now you can compile:





mkdir build
cd build
cmake ..
make





Eventually, you can run the executable:





> ./implied_vol
Implied Volatility: 0.318886

									            

        


		            

				                    

						                            
												                            0 comments