Wednesday, January 15, 2025

Building Curves from GTR Data

 

M 917 536 3378

maksim_kozyarchuk@yahoo.com






Kupala-Nich Mission and GTR

The mission of Kupala-Nich is to provide portfolio valuation for free and in open source. While this is achievable for listed securities using market data from platforms like Yahoo Finance, the challenge lies in valuing over-the-counter (OTC) derivatives such as swaps, forwards, and options. Unlike listed securities, OTC derivatives require both models and market data for valuation. While models for vanilla derivative products are relatively standardized and well-supported in the open-source space, market data is a different story. Not only is it unavailable for free, but market data vendors often charge exorbitant fees, with pricing structures designed to monetize every user who accesses this data. Such a model is inherently unscalable for a platform like Kupala-Nich.

GTR to the rescue. In response to the 2007 financial crisis, the Dodd-Frank Act established Swap Data Repositories (SDRs) and Global Trade Repositories (GTRs) to enhance transparency in OTC derivatives markets. The CFTC section of the GTR hosted on DTCC provides open access to anonymized records of all OTC trades conducted within U.S. jurisdiction. This data, accessible within an hour of trade execution through the CFTC Dashboard, reflects current trading levels for a wide range of instruments, including interest rate swaps, FX forwards, and options. This data provides an accurate representation of the market both intraday and at EOD, albeit with a slight delay.



How Can GTR Help with Derivative Valuation?

The fundamental building block of derivative valuation is discount and forward curves, which are required for pricing most, if not all, derivative products. By leveraging GTR, Kupala-Nich can extract market levels for Swap Rates across a broad range of currencies, giving us the ability to build both discount and forward curves.

The CFTC section of GTR has five sections: Commodities, Credits, Equities, FX, and Rates. While all tabs will be useful over time, to build curves, we will use the Rates section for G10 currencies and the combination of Rates and FX sections for NDF currencies. Below are daily counts of IR Swap and FRA trades reported on GTR for major currencies. As you can see, the volumes are substantial enough to provide indicative levels across a broad range of currencies. FX volumes are similar, though weighted more toward NDF currencies as deliverables are generally not reportable.

Currency

2025-01-06

2025-01-07

2025-01-08

2025-01-09

2025-01-10

2025-01-13

EUR

3656

14214

4621

8246

4372

3734

USD

5403

5952

6236

4002

6270

5734

GBP

1249

1225

1642

1561

1389

1480

JPY

1049

974

1196

883

1197

89

AUD

1438

822

986

440

447

1122

MXN

621

870

737

872

824

525

CAD

572

843

698

554

798

971

CHF

372

615

493

440

181

369

CLP

527

407

595

285

296

277

SEK

43

377

917

220

445

206

CZK

160

190

1046

150

181

259

ZAR

227

301

291

189

304

308

INR

213

340

329

235

215

150

KRW

175

227

166

196

225

269

BRL

149

247

323

120

167

201

CNY

251

122

150

93

365

201

SGD

159

260

304

113

170

140



Building the USD-SOFR-OIS Curve

As a starting point and to validate GTR as a reliable source for curve construction, the focus is on the USD-SOFR-OIS Curve. Using GTR, the curve is constructed from available trade data. For example, on 2025-01-13, there were 5734 trades available as inputs for the curve. Narrowing down to IR Swaps on USD-SOFR-COMPOUND yields 4400 trades. The majority of these are either spot starting or starting on the next IMM date. IMM-starting OIS Swaps often have upfront fees, making them more challenging for curve calibration, so the focus is on spot-starting OIS swaps, leaving 1700 trades.

Further filtering for non-standard terms and excluding maturities with only one trade leaves approximately 500 trades with the following distribution:


Date

Count

Notional

min

max

avg

median

2025-02-15

11

14,100,000,000

0.04305

0.04308

0.04306

0.04306

2025-03-15

2

950,000,000

0.04311

0.04312

0.04311

0.04311

2025-04-15

6

2,122,000,000

0.04307

0.04310

0.04308

0.04308

2025-05-12

3

2,070,000,000

0.04304

0.04307

0.04306

0.04307

2025-05-15

3

900,000,000

0.04304

0.04306

0.04305

0.04306

2025-07-15

7

642,800,000

0.04292

0.04298

0.04295

0.04296

2025-10-15

4

331,000,000

0.04279

0.04286

0.04283

0.04283

2025-12-15

4

211,000,000

0.04277

0.04291

0.04287

0.04290

2026-01-15

39

6,687,550,000

0.04040

0.04296

0.04273

0.04277

2026-07-15

3

77,000,000

0.04226

0.04239

0.04234

0.04236

2026-09-30

2

600,000,000

0.04227

0.04228

0.04227

0.04227

2026-12-31

2

250,000,000

0.04264

0.04266

0.04265

0.04265

2027-01-15

45

3,760,310,000

0.04237

0.04271

0.04254

0.04255

2028-01-15

27

954,000,000

0.04246

0.04282

0.04263

0.04263

2029-01-15

16

740,000,000

0.04255

0.04288

0.04276

0.04279

2030-01-15

69

3,351,000,000

0.04257

0.04300

0.04283

0.04286

2031-01-15

9

314,000,000

0.04279

0.04305

0.04291

0.04290

2032-01-15

20

1,212,000,000

0.04260

0.04314

0.04286

0.04288

2033-01-15

3

72,000,000

0.04285

0.04306

0.04296

0.04296

2034-01-15

2

6,000,000

0.04299

0.04304

0.04301

0.04301

2035-01-15

119

2,355,000,000

0.04268

0.04329

0.04298

0.04298

2037-01-15

5

30,000,000

0.04284

0.04327

0.04313

0.04319

2038-01-15

2

34,000,000

0.04325

0.04327

0.04326

0.04326

2039-01-15

2

22,000,000

0.04331

0.04332

0.04332

0.04332

2040-01-15

9

72,000,000

0.04320

0.04345

0.04334

0.04335

2045-01-15

12

243,000,000

0.04275

0.04337

0.04299

0.04295

2050-01-15

6

52,000,000

0.04207

0.04241

0.04228

0.04231

2055-01-15

47

596,900,000

0.04095

0.04153

0.04124

0.04125


Plotting the average and median rates yields the following:


Observations:

  • The median appears to offer a smoother curve than the average.

  • There are more points than a typical curve would have.

  • Volume is clustered around 1m, 3m, 6m, 1y, 2y, 3y, 5y, 7y, 10y, 20y, and 30y points.

  • There are some outliers around the 18-month point.

Using 1m, 3m, 6m, 1y, 2y, 3y, 5y, 7y, 10y, 20y, and 30y buckets and rolling up dates around them yields a much smoother curve. Plotting six days' worth of dates side by side confirms this approach. This curve looks reasonable and consistent.

The curve built from GTR data is not a point-in-time curve but rather a reflection of trades executed over a period. This analysis is based on daily trades, but for intraday curves, the sampling range could be reduced to several hours to provide a more real-time view.



Typically, OIS curves are built using Bloomberg tickers. These tickers are based on polling of dealer quotes conducted by Bloomberg and should provide a more consistent point-in-time snapshot. GTR curves, on the other hand, use historical trades to imply fair value rates. If a historical time series of Bloomberg quotes becomes available, a historical regression analysis can be performed to compare OIS curves derived from Bloomberg vs. GTR.

Below are some commonly used Bloomberg tickers for USD OIS:

  • USD 1 week: USOSFR1Z BGN Curncy

  • USD 1 month: USOSFRA BGN Curncy

  • USD 3 months: USOSFRC BGN Curncy

  • USD 6 months: USOSFRF BGN Curncy

  • USD 9 months: USOSFRI BGN Curncy

  • USD 1 year: USOSFR1 BGN Curncy

  • USD 2 years: USOSFR2Y BGN Curncy

  • USD 3 years: USOSFR3Y BGN Curncy

  • USD 5 years: USOSFR5Y BGN Curncy

  • USD 7 years: USOSFR7Y BGN Curncy

  • USD 10 years: USOSFR10Y BGN Curncy

  • USD 20 years: USOSFR20Y BGN Curncy

  • USD 30 years: USOSFR30Y BGN Curncy

By using these Bloomberg tickers for validation, the derived GTR curves can be compared against industry-standard benchmarks to ensure accuracy and reliability.



Next Steps

GTR data looks quite promising, but there is still a lot of work ahead on choosing a modeling library, designing data models for curve and swap repositories, as well as building UIs to visualize curves and to drill into and explain swap valuations. If you have any opinions or suggestions, I would love to hear from you!