Continuous Futures Backtesting: Choosing the Right Price Data

A continuous futures chart rises from 98 to 102, a gain of about 4.08%. Yet in the hypothetical trades below, an investor holding ten long contracts in each contract month finishes with zero futures profit. The difference is not a fee deduction. Part of the chart’s rise comes from changing the contract being displayed.

A futures backtest needs one set of rules for generating signals and another for reconstructing trades. Additive and ratio adjustments preserve different features of the price history. Neither replaces the original contracts, position sizes and execution records needed to calculate profit and loss.

Conceptual illustration of contract rolling and a position ledger, not a chart of market prices or calculated returns.

A 4.08% chart gain, but no profit on ten contracts

Consider a hypothetical commodity with nearer-dated contract A and later-dated contract B. Both are quoted in US dollars per unit, with 1,000 units per contract. A starts at 98. At the roll, A is quoted at 100 and B at 104 at the same observation time. B subsequently falls to 102.

Assume comparable price sampling and, separately, that the stated trades can execute at those prices. The base calculation excludes commissions, bid–ask spreads, slippage, cash interest and funding constraints. It retains the four-point calendar spread between A and B. Buy ten A, close them at the roll, open ten B and later close B:

A chart can instead connect 98, 100, 104 and 102. Its endpoint calculation, 102 ÷ 98 − 1, gives approximately 4.08%. But the jump from 100 to 104 switches the quoted instrument; it is not another four-point gain on A. The investor never held one contract all the way from 98 to 102.

Nor is there an additional $40,000 loss when B replaces A. B is opened and valued at 104, so its initial P&L is zero. Subtracting the calendar spread again after calculating both holding periods would turn a difference between contract prices into a cash debit that these trades did not incur. Futures are generally marked to market in cash daily; the table aggregates each position’s cumulative gains and losses.

If B instead ends at 108, its profit is $40,000 and the combined profit becomes $60,000. A more expensive deferred contract therefore does not make a loss inevitable. Why it trades at a premium is a question about futures basis and term structure; the holding result depends on the actual contracts’ subsequent price paths.

What additive and ratio adjustments preserve

Continuous data can leave the roll gap intact or adjust historical prices to close it. Both adjustment methods below keep the latest B prices unchanged and rewrite only the earlier A segment. Names vary across data systems, so establish the anchor and formula rather than relying on an adjustment label.

The two roll rows are simultaneous quotations, not successive market moves. Additive adjustment preserves points within a uniformly adjusted segment. Adding 104 − 100 = 4 changes A’s rise from 98–100 to 102–104. It remains two points, but the percentage move falls from about 2.04% to 1.96%. Keeping B at 104 while subtracting four from A at 100 would produce 96 and widen the gap to eight points: the wrong direction for this anchor.

Ratio adjustment preserves percentages within a segment receiving the same multiplier. Multiplying A by 104 ÷ 100 = 1.04 changes 98 to 101.92 and 100 to 104. The segment still rises about 2.04%, but now by 2.08 points. The full ratio series rises about 0.0785% from 101.92 to 102, still different from the fixed-ten-contract result.

These are same-segment properties. Once a calculation window includes both adjusted A and unadjusted B, recalculate it using each observation’s actual factor. A relationship preserved inside A does not automatically preserve a moving average or any other indicator spanning the roll.

A 2% trigger shows why the data choice changes the signal

Suppose a signal requires a gain of at least 2% over the A segment. Original prices give 100 ÷ 98 − 1 ≈ 2.0408%, so the condition passes. Ratio-adjusted prices give 104 ÷ 101.92 − 1 ≈ 2.0408%, also a pass. Additively adjusted prices give 104 ÷ 102 − 1 ≈ 1.9608%, so the condition fails.

The hypothetical market has not changed, but the input to the rule has. Adjustment can therefore change whether a signal appears, not merely the appearance of a chart. This comparison tests only the A segment under one common adjustment. It does not establish that a cross-roll signal, or the complete strategy, remains unchanged.

Consider a separate order scenario: A has already reached 100 when a trader sets a long-position stop at 99. A later four-point adjustment places that coordinate at 103 on the chart. The original order remains at 99. Test prices after the order became active on contract A, together with the execution assumptions. A crossing of 103 on the adjusted curve is not enough to record a fill.

For a cross-roll indicator, map every observation’s date, actual contract, original price, adjustment factor and adjusted price. Rebuild the signal, then simulate orders in the actual contract. Zero and negative prices require explicit treatment: division by zero is undefined, a negative price has no real logarithm, and percentage changes across zero can lose their usual return interpretation. A negative continuous value can also come from accumulated additive adjustments rather than an original negative quotation; inspect the underlying contract history.

Correct signal data still leaves a position-sizing decision

The zero-profit result assumed ten contracts in both holding periods. Now target $980,000 of notional at each entry, keep the quantity unchanged within each holding period, and do not add A’s profit to B’s target.

A still requires $980,000 ÷ (98 × 1,000) = 10 contracts. B requires $980,000 ÷ (104 × 1,000) ≈ 9.4231 contracts. Using unrounded quantities, A earns $20,000 and B loses about $18,846.15. The combined profit is approximately $1,153.85 because fewer B contracts are exposed to the decline.

Fractional quantities isolate the sizing rule mathematically; they are not assumed tradable in a standard contract. Implementation must handle whole contracts, smaller contract sizes where available, residual cash, margin and execution differences. The $980,000 target is neither a full commodity purchase payment nor required margin. If the strategy is separately assigned $980,000 of initial capital with no external deposits or withdrawals, its futures P&L represents about 0.1177% of that capital. A different capital denominator gives a different percentage.

Conversely, ten contracts keep the count fixed but move notional exposure from $1,000,000 immediately before the roll to $1,040,000 afterward. The 4% rise does not establish an identical change in margin or risk. Actual account results also need actual cash income and costs. In a separate scenario, zero futures P&L plus $1,200 of eligible cash interest less $300 of all relevant trading costs gives $900 net profit. Do not assume the entire notional earns interest, or deduct the calendar spread again.

Check the underlying contracts before accepting the backtest

First identify what each date represents. An active-contract series needs a rule for selecting the active month; a front-month series needs a rule for leaving the expiring contract. Neither is a single standard contract that never expires. Obtain actual contract identifiers, roll timestamps and price-sampling conventions. An official settlement price is not the same thing as the last traded price, and comparable quotations alone do not establish executable fills.

Then take at least two different original prices from one old-contract segment and compare them with the chart. Original prices of 98 and 100 appearing as 102 and 104 have a common difference of four, consistent with additive adjustment. If they appear as 101.92 and 104, both ratios to the original prices are 1.04, consistent with ratio adjustment. One point cannot distinguish those explanations. Two points test only the selected segment, not the method for the entire series.

Keep signal verification separate from execution verification. Match each proposed trade to its contract, direction, timestamp, price, quantity, multiplier and costs. A particular FastBull series still requires its own construction rules and mapping; this illustration does not establish the platform’s algorithm. Buy and sell markers on a continuous chart, without the original contracts, are insufficient evidence that those trades could have occurred.

Finally, check when the inputs became available. A day’s final volume cannot select the active contract at that same day’s opening. Later adjustments that rewrite history also need signal-by-signal testing: fixed price thresholds may change, while some differences within a uniformly shifted segment may not. If contract identities, adjustment rules or quantities are missing, fill those gaps first. A smoother line is not a substitute for the missing evidence.