Winservers.NET

StrategyQuant X · Robustness checks

How Long Do Walk Forward and Monte Carlo Checks Take in StrategyQuant X?

Generation speed is what people compare. Robustness checks are what they wait for. Here is the measured time of every check from five weeks of real StrategyQuant X work on one server.

Quick answer: on a 16-thread server with about 7 years of data, Walk Forward took about 98 seconds per strategy, Randomize Data about 136 seconds and SPP about 39 seconds. OOS and trade manipulation took under a second. One thousand candidates through Randomize Data alone is about 37.7 h.

Measured

Time per strategy for every check

The run lasted from August 4 to September 9, 2026: 15 chained projects (the pipeline step by step), genetic generation, MetaTrader 5 engine, timeframes M30 to H4. Server: Intel Core i9-13900 with 16 threads in use (15 for SQX), 64 GB DDR5, NVMe, StrategyQuant X build 144. All figures are calculated from the SQX project logs.

CheckStrategies checkedPass rateTotal hoursShare of check timeSeconds per strategy
OOS (out of sample)66,69176.6%3.20.6%0.2
Trade Manipulation7,27085.1%0.60.1%0.3
Tick Retest5,72980.1%12.62.5%7.9
SPP (System Parameter Permutation)6,93682.2%75.715.1%39
Walk Forward5,50683.4%149.629.8%98
Randomize Data (Monte Carlo)6,90080.8%260.251.8%136

Generation took 190.6 hours. The checks took 502 hours, 2.6 times more. Randomize Data and Walk Forward together used 82% of all check time.

About settings: time per strategy depends on how each check is configured: the number of Monte Carlo simulations, parameter permutations and Walk Forward runs. The SQX logs record durations and counts, not these settings, so treat the figures as one real configuration and not as a universal constant. Counts are passed plus failed, because the Total field in the retest logs is not filled correctly.

Why the gap

Why some checks take a second and others two minutes

Cheap: OOS and trade manipulation

The OOS check looks at the out-of-sample part of a backtest. Monte Carlo trade manipulation shuffles or skips trades in the existing list of trades. Neither needs many new backtests, so both stay under a second per strategy.

Medium: tick retest

One more backtest, but on higher precision data. About 7.9 seconds per strategy here.

Expensive: SPP and Randomize Data

System Parameter Permutation backtests the strategy with many parameter combinations. Randomize Data backtests it on many altered versions of the price history. Each simulation is a full new backtest, so time grows with the number of simulations.

Expensive: Walk Forward

Walk Forward re-optimizes the strategy on a rolling window and tests each step on the data that follows. That is many optimizations per strategy. A Walk Forward Matrix repeats this for a grid of settings and takes proportionally longer.

Finding 1

A higher timeframe does not rescue the checks

In generation, timeframe decides everything: H4 was 8.5 times faster than M30. In the checks it barely mattered.

Check, seconds per strategyM30H1H2H4
Tick Retest8.57.68.67.4
SPP (System Parameter Permutation)37413541
Walk Forward9410210185
Randomize Data (Monte Carlo)131144116146

Walk Forward stayed between 85 and 102 seconds per strategy, Randomize Data between 116 and 146. We did not isolate the reason, so we report it as an observation from this run. The practical point: moving research to H4 will speed up the Builder much more than it will speed up validation.

Finding 2

The average hides a long tail

Strategies differ a lot in how long they take to check. Most batches are quick, a few are very slow, and the slow ones set the calendar time.

CheckAverageMedian runSlowest 10% of runs, fromLongest single task
Tick Retest7.9 s4.9 s29 s14 min
SPP (System Parameter Permutation)39 s21 s97 s2.7 h
Walk Forward98 s19 s307 s7.7 h
Randomize Data (Monte Carlo)136 s89 s297 s7.8 h

For Walk Forward the median run needed 19 seconds per strategy while the average was 98. The longest single Walk Forward task ran 7.7 hours and the longest Randomize Data task 7.8 hours. A typical batch in this pipeline was small: about 7 strategies per check.

Plan with the average, expect the tail. If you estimate from a few quick strategies, you will underestimate. Leave room for single tasks that run for hours, and use a machine that will not be switched off in the middle.

Your turn

How to estimate your own run

Multiply the number of strategies that enter a check by its seconds per strategy. With the averages from this run, on the same server:

Strategies entering the checkTick RetestSPPWalk ForwardRandomize Data
10013 min1.1 h2.7 h3.8 h
5001.1 h5.5 h13.6 h18.9 h
1,0002.2 h10.9 h27.2 h37.7 h
5,00011.0 h2.3 days5.7 days7.9 days

To use your own numbers, open user\projects\<your project>\log\global_log_*.log and find the line with Time per strategy after each AutomaticRetest task. Compare projects with the same check settings only.

Put cheap checks first

Every check removes part of the candidates. If the cheap ones run first, fewer strategies reach the expensive ones. An illustration with the pass rates and times above, treating the pass rates as independent:

Order of the four slow checksTotal time for 1,000 candidates
Expensive first: Randomize Data, Walk Forward, SPP, Tick Retest2.8 days
Cheap first: Tick Retest, SPP, Walk Forward, Randomize Data2.1 days

Same checks, same result set, about 27% less time. In a real project pass rates are not independent, so the saving will differ, but the direction holds.

What shortens validation

  • Cheap checks and filters before expensive ones.
  • Fewer simulations where a smaller number still answers the question.
  • Stricter Builder filters, so fewer weak candidates arrive.
  • More CPU threads: every check is parallel backtesting.

What does not

  • Switching to a higher timeframe and expecting the 8x gain of the Builder.
  • Estimating from the first few fast strategies.
  • Running checks on a PC that sleeps or restarts at night.
  • Sharing the machine with live trading terminals.

Hardware

What this means for the server

In this run the checks took 72% of all working time. A server for StrategyQuant X should be sized for Walk Forward and Monte Carlo, not for the Builder alone: many real threads, enough RAM for long runs and the ability to work for weeks without interruption.

Official and measured requirements are in StrategyQuant X system requirements. Generation speed by timeframe is in the StrategyQuant X benchmark, and the path from 130.7 million generated strategies to 669 final ones is in How many strategies do you need to generate? The full budget is in how much StrategyQuant X costs. Servers and prices are on the dedicated servers page.

Ask which server fits my SQX checks

FAQ

Frequently asked questions

How long does Walk Forward take in StrategyQuant X?

In our measured run Walk Forward took about 98 seconds per strategy on average, on a 16-thread Intel Core i9-13900 server with about 7 years of data. Single runs varied a lot: the median was 19 seconds and the slowest 10% took more than 307 seconds per strategy. Your time depends on the number of Walk Forward runs and optimization steps you set.

How long does a Monte Carlo test take in StrategyQuant X?

It depends on the type. Monte Carlo trade manipulation works on the list of trades and took about 0.3 seconds per strategy. Randomize Data needs a new backtest for every simulation and took about 136 seconds per strategy in our run.

Which StrategyQuant X robustness test is the slowest?

Randomize Data, then Walk Forward. Together they used 410 of 502 hours of check time in our five-week run, about 82%.

Does a higher timeframe make robustness checks faster?

Much less than it speeds up generation. In our run generation on H4 was 8.5 times faster than on M30, but Walk Forward took 85 to 102 seconds per strategy and Randomize Data 116 to 146 seconds on every timeframe.

How can I shorten robustness testing?

Run cheap checks first so fewer strategies reach the expensive ones, reduce the number of simulations where a smaller number still answers your question, and filter harder before the checks. More CPU threads shorten every check.

Do passed robustness checks mean the strategy will be profitable?

No. A passed check only means the strategy survived that test on historical data. It does not predict live results. Winservers.NET rents servers and does not give trading advice.

Waiting days for Walk Forward or Monte Carlo?

Send us your check settings, timeframe and the number of candidates per day. We will suggest a dedicated server for StrategyQuant X and tell you what to expect from it.

Ask in WhatsApp or Telegram

K.S.M. Trade Sp. z.o.o. | NIP/VAT PL5273128546 | REGON 52957272900000 | Warsaw, Poland