Stochastic Spread Pairs Trading in the Indian Commodity Market

Abhijeet Chandra; Dhruv Mahajan

arxiv: 1907.08397 · v1 · pith:4W7E7AEJnew · submitted 2019-07-19 · 💱 q-fin.TR

Stochastic Spread Pairs Trading in the Indian Commodity Market

Dhruv Mahajan , Abhijeet Chandra This is my paper

Pith reviewed 2026-05-24 19:08 UTC · model grok-4.3

classification 💱 q-fin.TR

keywords pairs tradingstochastic spreadcointegrationIndian commoditiesSharpe ratioMCX spot pricesbacktestingdifferential evolution

0 comments

The pith

A stochastic spread model on cointegrated Indian commodity pairs delivers Sharpe ratios above 1.4 in backtesting.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The authors apply Johansen cointegration tests to spot prices of 17 Indian commodities from 2010 to 2018 and identify 12 tradable pairs. They fit a single-factor stochastic model to the logarithmic spread of each pair, estimate parameters with differential evolution, and optimize trading rules on an 80 percent training window. The same rules applied to the held-out 2017-2018 period produce Sharpe ratios above 1.4 for every pair. A sympathetic reader would care because the result indicates that mean-reverting spread dynamics can be exploited for risk-adjusted profit in an emerging-market commodity setting without requiring the underlying prices themselves to revert.

Core claim

The paper shows that a single-factor stochastic trading approach applied to the logarithmic spread of 12 cointegrated commodity pairs, with parameters estimated by differential evolution and trading thresholds optimized by backtesting on 2010-early 2017 data, produces Sharpe ratios above 1.4 when the identical rules are run on the 2017-2018 out-of-sample window.

What carries the argument

Single-factor stochastic model on the logarithmic spread of Johansen-cointegrated pairs, with parameters fitted by differential evolution and entry-exit thresholds chosen by training-period backtesting.

If this is right

All twelve identified pairs generate positive risk-adjusted returns under the stochastic spread rule.
The differential-evolution parameter estimates produce stable trading signals across energy, metal, and agricultural sectors.
An 80:20 chronological split suffices to validate the strategy within the 2010-2018 window.
The logarithmic spread formulation captures the long-run relationship sufficiently for profitable mean-reversion trades.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The same modeling pipeline could be rerun on futures prices to test whether basis risk alters the observed Sharpe ratios.
If cointegration breaks in a later regime, retraining the model on rolling windows would be required to maintain performance.
The approach leaves open whether adding a second stochastic factor or volatility filter would further improve drawdown control.

Load-bearing premise

The cointegration relationships found in the training data remain valid in the testing period and the trading-rule parameters chosen on training data continue to work out of sample.

What would settle it

Applying the identical model and thresholds to MCX spot data from 2019 onward and finding that the Sharpe ratio for the same 12 pairs falls below 0.5 or turns negative.

read the original abstract

In this study, we applied a stochastic spread pairs trading strategy on the Indian commodity market. The complete set of commodities were taken whose spot price was available for the period of January 1st 2010 to December 31st 2018 including energy, metals and the agricultural commodity sector. Spot data was taken from the MCX pooled spot prices for 17 commodities. The data was split into training period (January 1st 2010 to 14th March 2017) and testing period(15th Match 2017 to 31st December 2018). The splitting was done using a 80:20 split.Johanssen Cointegration tests were done on training data for pairs of commodities to check for long-run relationship and the cointegrated commodities were selected for formation of the trading process. We found a total of 12 cointegrated pairs out of 136 possible pairs. Cointegration was assumed for the testing period. A single-factor stochastic trading approach was applied on the logarithmic spread of the cointegrated pairs for both the training and testing period.The parameters of stochastic spread model were estimated using differential evolution algorithm. Also parameters for the trading rule were optimized by backtesting on the training period and assumed for the testing period. The results show a sharpe ratio of above 1.4 for all the commodity cointegrated pairs in the backtesing period.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

Sharpe ratios above 1.4 appear only on the training period after explicit parameter optimization there; the paper splits the data but supplies no test-period numbers at all.

read the letter

The paper runs Johansen cointegration tests on 17 MCX commodities over 2010-2018, identifies 12 pairs on the 2010-early 2017 training window, fits a single-factor stochastic spread model via differential evolution, tunes trading thresholds by backtesting on that same window, and reports Sharpe ratios above 1.4 for those pairs in the backtesting period. Cointegration and parameters are simply assumed to carry over to the 2017-2018 test window with no further checks or results shown there. The methods themselves are standard applications of existing tools to a new dataset; nothing in the derivation or modeling step is new. The work is clear about what it did and cites the usual references for cointegration and stochastic spreads. The central limitation is exactly the one the stress-test note flags: performance is measured only where the parameters were chosen, so the numbers do not speak to whether the relationships or rules survive the regime the authors themselves set aside for testing. A reader who wants a descriptive list of which Indian commodity pairs showed cointegration in that sample might find the pair count useful, but anyone looking for evidence that the strategy produces reliable risk-adjusted returns will not get it from the reported figures. This does not rise to the level that would justify sending it to referees.

Referee Report

3 major / 2 minor

Summary. The paper applies a stochastic spread pairs trading strategy to 17 Indian commodities using MCX spot prices (2010-2018). Data is split 80:20 into training (1 Jan 2010–14 Mar 2017) and testing (15 Mar 2017–31 Dec 2018) periods. Johansen cointegration tests on training data identify 12 cointegrated pairs. A single-factor stochastic model is fit to the log spread via differential evolution; trading-rule parameters are optimized by backtesting on training data and assumed for testing. Cointegration is assumed to persist. The sole performance claim is Sharpe ratios above 1.4 for all pairs in the backtesting (training) period.

Significance. If the strategy had produced verified out-of-sample Sharpe ratios above 1.4 after parameter optimization, it would offer evidence for a viable stochastic-spread pairs trading approach in Indian commodities. As presented, however, the results are in-sample after explicit optimization on the evaluation data, so the manuscript provides no basis for assessing significance or generalizability.

major comments (3)

[Abstract] Abstract: The central claim ('sharpe ratio of above 1.4 for all the commodity cointegrated pairs in the backtesing period') is evaluated exclusively on the training period, where both stochastic spread model parameters (via differential evolution) and trading thresholds were optimized by backtesting. No Sharpe ratios, returns, or performance metrics are supplied for the testing period despite the explicit 80:20 split and the assumption that cointegration persists.
[Abstract] Abstract and Results: Trading-rule parameters were chosen by backtesting on the training period and the reported Sharpe ratios are also computed on that same training period, so the performance metric is constructed from quantities fitted to the evaluation data. This circularity means the figures do not test whether the identified relationships or chosen thresholds survive the regime shift implied by the split.
[Methods] Methods/Results: No out-of-sample validation, error bars, robustness checks, or even confirmation that cointegration holds are mentioned for the testing window (15 Mar 2017–31 Dec 2018). The assumption of persistence is stated but never verified against the held-out data.

minor comments (2)

[Abstract] Abstract: Typo 'backtesing' should read 'backtesting'; '15th Match 2017' should read '15th March 2017'.
[Abstract] Abstract: The text states the strategy is applied 'for both the training and testing period' yet supplies no results or even basic statistics for the testing period.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the detailed and constructive report. The comments accurately identify that the manuscript reports performance only on the training period after explicit optimization of both model parameters and trading thresholds, with no results shown for the held-out testing period. We agree this constitutes a significant limitation for evaluating the strategy. We will revise the manuscript to compute and report out-of-sample trading performance on the testing window using parameters fixed from the training data, verify cointegration persistence, and update the abstract and results sections accordingly.

read point-by-point responses

Referee: [Abstract] Abstract: The central claim ('sharpe ratio of above 1.4 for all the commodity cointegrated pairs in the backtesing period') is evaluated exclusively on the training period, where both stochastic spread model parameters (via differential evolution) and trading thresholds were optimized by backtesting. No Sharpe ratios, returns, or performance metrics are supplied for the testing period despite the explicit 80:20 split and the assumption that cointegration persists.

Authors: We agree that the abstract and results report the Sharpe ratio claim exclusively from the training period. The manuscript does not supply any performance metrics for the testing period. In revision we will add the out-of-sample Sharpe ratios, returns, and other metrics computed on 15 Mar 2017–31 Dec 2018 with parameters held fixed from the training period. revision: yes
Referee: [Abstract] Abstract and Results: Trading-rule parameters were chosen by backtesting on the training period and the reported Sharpe ratios are also computed on that same training period, so the performance metric is constructed from quantities fitted to the evaluation data. This circularity means the figures do not test whether the identified relationships or chosen thresholds survive the regime shift implied by the split.

Authors: The referee correctly notes the in-sample optimization and evaluation. This circularity is a genuine limitation of the presented results. We will address it by reporting performance on the testing period using the training-optimized parameters, thereby testing survival across the split. revision: yes
Referee: [Methods] Methods/Results: No out-of-sample validation, error bars, robustness checks, or even confirmation that cointegration holds are mentioned for the testing window (15 Mar 2017–31 Dec 2018). The assumption of persistence is stated but never verified against the held-out data.

Authors: We acknowledge that the current text provides no out-of-sample validation or cointegration check on the testing window. We will add verification of cointegration on the held-out data, the corresponding trading results with the fixed parameters, and any applicable robustness metrics. revision: yes

Circularity Check

1 steps flagged

Sharpe ratios >1.4 reported only on training/backtesting period after explicit parameter optimization; no out-of-sample results supplied for held-out testing window

specific steps

fitted input called prediction [Abstract]
"Also parameters for the trading rule were optimized by backtesting on the training period and assumed for the testing period. The results show a sharpe ratio of above 1.4 for all the commodity cointegrated pairs in the backtesing period."

Parameters are fitted via backtesting on the training data; the sole reported performance figure (Sharpe >1.4) is then computed on that identical training/backtesting window, so the metric is constructed from quantities optimized to the evaluation data rather than tested out-of-sample.

full rationale

The paper explicitly optimizes trading-rule parameters by backtesting on the training window and then reports the central performance metric (Sharpe >1.4) exclusively on that same backtesting/training period. Although a testing period is defined and cointegration is 'assumed' to hold there, no Sharpe ratios, returns, or even confirmation of cointegration persistence are provided for 15 Mar 2017–31 Dec 2018. The reported results therefore reduce to in-sample performance after fitting, violating the claim of a predictive or out-of-sample strategy evaluation.

Axiom & Free-Parameter Ledger

2 free parameters · 2 axioms · 0 invented entities

The central claim rests on fitted stochastic-model parameters and trading thresholds chosen to maximize in-sample performance, plus the untested assumption that cointegration persists out-of-sample.

free parameters (2)

stochastic spread model parameters
Estimated via differential evolution on training data
trading rule thresholds
Optimized by backtesting on training data

axioms (2)

domain assumption Cointegration relationships identified on training data persist into the test period
Explicitly assumed after testing only on training data
standard math Johansen test correctly identifies tradable long-run relationships
Standard assumption invoked when selecting the 12 pairs

pith-pipeline@v0.9.0 · 5781 in / 1466 out tokens · 27777 ms · 2026-05-24T19:08:10.466354+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

A single-factor stochastic trading approach was applied on the logarithmic spread... parameters... optimized by backtesting on the training period

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.