Robust Multi-Asset Trend Following in Global Futures
Time-series momentum, volatility normalization, trading costs, diversification, and robustness across 75 global futures — the first completed study in this research program.
Research question
Do systematic futures strategies remain credible after the assumptions that usually make a backtest look attractive — contract economics, currency conversion, volatility forecasting, whole-contract sizing, costs, concentration limits, and leverage — are made explicit, and can economically distinct, pre-specified strategies improve portfolio risk without retrospective selection?
Why this matters
A strong historical equity curve can depend on continuous-contract construction, incomplete cost assumptions, unrealistic sizing, excessive leverage, or a few unusually profitable markets. This study asks what remains of a literature-based trend benchmark after each of those conveniences is removed, and freezes the specification for prospective monitoring so the next evidence is genuinely out of sample.
Hypotheses
- 01H1: A literature-based 12-month time-series momentum rule retains positive net performance across a frozen 75-market futures universe after realistic commissions, slippage, and roll execution.
- 02H2: The result is broad-based across markets, asset classes, and time rather than concentrated in a few contracts or years.
- 03H3: Economically distinct, pre-specified strategies (breakout, cross-sectional momentum, term structure) remain positive after costs but do not displace the time-series momentum benchmark as the strongest standalone sleeve.
- 04H4: A fixed equal-risk combination of the four sleeves reduces drawdown without sacrificing Sharpe, and locked portfolio controls trade a bounded amount of return for a materially more implementable account.
Methodology
Baseline signal
A 12-month / 252-trading-bar time-series momentum sign: long when the current price exceeds its 252-bar lag, short when below. Norgate back-adjusted continuous series drive signal research; unadjusted series and individual contracts are retained for implementation and roll checks.
Volatility estimation and sizing
A lagged EWMA of daily point changes with a 60-trading-day center of mass, annualized with 252 days, scales each instrument to a common ex-ante volatility convention. Positions round to whole contracts with a dollar-volatility floor for eligibility.
Transaction costs
USD 2.50 commission plus one tick of slippage per contract leg, with contract-specific tick values, currency conversion, and explicit roll activity. The final constrained account absorbs USD 2.93 million of modeled costs across 228,814 contract transaction legs in the baseline cost study.
Portfolio construction
Four pre-specified sleeves — 12-month time-series momentum, a 252-day entry / 126-day exit price-channel breakout, 12-minus-1-month cross-sectional momentum, and a constrained term-structure strategy — combine at fixed equal risk. Weights, caps, and constraints were locked before any portfolio result was observed.
Final controls
A risk target, per-market and asset-class caps, a leverage ceiling, a gross notional limit, and whole-contract rounding produce the implementable account. The controls bind economically: the leverage ceiling constrained 56 monthly decisions, the market cap 552 dates, the asset-class cap 4,769 dates, and the gross notional cap 2,617 dates.
Research governance
Universe and methodological choices are frozen before strategy performance is accepted; changes must be data- or implementation-motivated, never performance-motivated. Every accepted RealTest result is independently reconciled in Python, and negative results are retained when they answer a legitimate question.
Data
Data source
Norgate Data daily futures: individual contracts plus pre-built back-adjusted and unadjusted continuous-futures series. Licensed vendor data is not redistributed.
Universe construction
The Core 75 universe was frozen without reference to strategy returns, after a metadata and history audit of the Norgate continuous-futures inventory.
Contract linkage validation
806 usable contract changes across E-mini S&P 500, 10-Year U.S. Treasury Note, Euro FX, and WTI Crude Oil (1990–2026) were reconciled event by event; the full-universe audit covers 13,661 vendor roll events with no duplicate market-date observations.
Term-structure treatment
The outgoing-to-incoming price difference is treated as a term-structure observation, not a transaction cost — the April 2020 WTI dislocation alone stood at USD 58,060 per contract. Execution cost remains limited to commissions and slippage on the two legs.
Key results
No results published yet
TSMOM sleeve (1991–2026)
—
10.32% CAGR, 0.964 Sharpe, −20.01% max drawdown
Breakout sleeve
—
7.54% CAGR, 0.717 Sharpe, −27.76% max drawdown
Fixed equal-risk portfolio
—
0.964 Sharpe, −14.94% max drawdown (25.3% shallower)
Constrained account
—
6.44% net CAGR, 7.19% volatility, 0.877 Sharpe, −11.95% max drawdown
Baseline cost impact
—
Gross 6.28% CAGR → net 4.46%; USD 2.93M modeled costs over 228,814 legs
Rolling five-year windows
—
31 of 31 positive net growth and Sharpe (overlapping, not independent)
Planned exhibits
- Baseline time-series momentum, gross and net
- Signal horizon and volatility-estimator research
- Transaction-cost and turnover attribution
- Breadth across markets, asset classes, and time
- Term-structure and roll-implementation evidence
- Standalone sleeve reconciliation
- Fixed equal-risk four-strategy portfolio
- Constrained implementable account
- Rolling five-year windows and prospective holdout
Every published chart will carry a descriptive title, axis labels, units, legend where needed, its sample period, a gross or net label, a short written interpretation, and accessible colors with tooltips.
Interpretation
What the evidence supports
A constrained portfolio of complementary systematic futures strategies under fixed weights. Every broad asset class contributed positively to the baseline: commodities 58.3% of net profit (29 of 31 markets profitable), government bonds 24.3%, equity indices 12.4%, and currencies 5.1%.
What it does not establish
Future profitability. The conclusion is deliberately narrower than the most attractive backtest: the rolling five-year windows overlap and are not independent trials, and the decisive next evidence comes from the untouched live ledger that begins after 14 August 2026.
Robustness checks
- Signal horizon comparison (3/6/12-month ensemble versus 12-month rule)
- Alternative volatility estimators
- Breakout parameter grid (all three pairs profitable, Sharpe 0.624–0.812)
- Cross-sectional momentum horizon sensitivity
- Neighboring-rule diagnostics around each accepted sleeve
- Government-bond cost and turnover diagnosis
- Full-universe roll-data validation (13,661 events)
- Rolling five-year vintages
- Prospective holdout with frozen specification
Limitations
- Licensed data cannot be redistributed, limiting direct third-party replication.
- Cost estimates are modeled commissions and slippage, not observed fills.
- Rolling five-year windows overlap and are not independent trials.
- Capacity and market impact are not modeled at institutional size.
- The term-structure sleeve loses money in government bonds after costs; the portfolio result depends on the fixed combination, not on every sleeve winning everywhere.
Conclusion
The evidence supports a constrained portfolio of complementary systematic futures strategies under fixed weights; it does not establish future profitability. Portfolio rules were fixed before combined results were examined, and no sleeve, weight, or constraint changed afterward. The final architecture is the outcome of a controlled research process, not the combination that happened to maximize historical performance. The specification is now frozen, and the prospective ledger that begins after 14 August 2026 is the decisive test.
Reproducibility
Audit trail
The report carries a full research audit trail (Appendix A): every numbered methodological decision, its motivation, and whether it passed. Contract metadata, roll inventories, and reconciliation outputs are retained as supporting diagnostics.
Independent reconciliation
Every accepted RealTest result — daily profit, costs, contract legs, market and asset-class contributions, rolls, FX conversion, and dates — is independently recomputed in Python before acceptance.
Data licensing
Norgate data is licensed and not redistributed. The documented schema, transformation steps, and validation audits allow reproduction with an equivalent authorized dataset.
Downloads and links
Citation
This is working research, not a peer-reviewed publication. If you refer to it, please cite it as work in progress and note the status shown above.
BibTeX
@misc{bang_robust_multi_asset_trend_2026,
author = {Bang, Pratik},
title = {Robust Multi-Asset Trend Following in Global Futures},
year = {2026},
note = {Status: Completed. Working research, subject to revision.},
url = {[PROJECT URL TO ADD]}
}Plain text
Bang, P. (2026). Robust Multi-Asset Trend Following in Global Futures. Working research (Completed). [PROJECT URL TO ADD]