Pick the right search
Which optimization method should you use?
TradingTune ships seven ways to search your parameter space, and the right one depends on two things: how big and bumpy the space is, and how many backtest cycles you can spend. This guide gives a short "use it when" call for each method, frames the trade-off as cycle budget versus thoroughness, and ends with a comparison table you can scan in one pass.
The real trade-off: cycle budget versus thoroughness
Every backtest cycle costs time, and on the free tier you get ten optimization cycles per day (a free account and sign-in are required to run). Pro lifts that to unlimited cycles per day. So the question behind every method choice is the same: given the cycles you can afford, how do you spend them to find strong settings without wasting runs?
At one end sits exhaustive search, which leaves nothing untested but only fits a handful of parameters before the grid explodes. At the other end sit learned and hybrid searches, which spend each cycle where the payoff looks most likely and so cover large spaces on a modest budget. Most of choosing a method is deciding where on that line your strategy sits.
Two quick heuristics. The smaller and smoother the space, the more a deterministic, exhaustive-leaning method earns its keep. The larger, bumpier, or more unfamiliar the space, the more a sample-efficient learner pays off per cycle. When in doubt, TradingTune already picks a sensible default for you based on how many parameters are enabled, and you can override it any time.
Use it when: a quick call for each method
Each card below names the single situation where that method is the strongest pick. Follow any name through to its method page for the full mechanics and defaults.
-
Use it when the grid is small and you want exhaustive certainty.
With three or fewer parameters over tight ranges, Brute Force tests every combination and provably returns the best one inside your ranges. There is no luck and no sampling. The catch is arithmetic: the cycle count is the product of the value counts, so it explodes the moment the space grows.
- 2
Random
Use it when you want a quick, broad read on a large or unfamiliar space.
Random search samples a fresh independent set every cycle, so it covers a wide space fast and never gets stuck. It is a surprisingly tough baseline and the honest control group: if a smarter method cannot beat random on your strategy, that tells you something about the landscape.
-
Use it when you are refining around an already-good configuration.
Sequential Improvements tweaks one parameter at a time in round-robin while holding the rest fixed. It is cheap and easy to read, ideal for probing one parameter's effect near a configuration you already trust. It is blind to interactions, so reach for it only when your parameters are roughly independent.
-
Use it when the landscape is rugged and you must escape local optima.
Simulated Annealing wanders widely early, then cools so late cycles only climb. That willingness to occasionally go downhill is what lets it escape a shallow dip toward a higher peak. It is the strong middle ground for four to ten parameters on a bumpy surface where a plain hill-climb keeps getting stuck.
-
Use it when the space is large and every cycle has to count.
TPE learns from every result, building a model of what works and spending cycles where the payoff is most likely. That sample efficiency is the best bang per cycle, so it shines on medium-to-large spaces where exhaustive search is hopeless and your budget is tight. It is the default for most strategies.
-
Use it when the metric responds smoothly to each parameter.
Bisection brackets the best region with a coarse sweep, narrows with golden-section search, then polishes the top values. It is fast, deterministic and resumable, ideal when the surface is smooth-ish and parameters are mostly independent. Rugged or strongly coupled landscapes are where you want the hybrid instead.
-
Use it when you want broad narrowing first, then a fine Bayesian polish.
Bisection then TPE Refine runs the full deterministic Bisection search, then hands the remaining budget to TPE to refine around the discovered peak. You get Bisection's speed on the broad search plus TPE's Bayesian polish near the optimum, in one run. It carries the highest effectiveness of the seven.
All seven methods at a glance
Cycle efficiency is how much useful progress each backtest tends to buy. Higher efficiency stretches a tight budget further; lower efficiency leans on raw cycle count.
| Method | Best for | Search style | Cycle efficiency |
|---|---|---|---|
| Brute Force | Tiny grids, definitive answers | Exhaustive sweep | Very low |
| Random | Fast first read, baselines | Uniform sampling | Low |
| Sequential Improvements | Independent parameters | One axis at a time | Low to medium |
| Simulated Annealing | Bumpy 4 to 10 parameter spaces | Cooling random walk | Medium |
| TPE (Bayesian) | Large spaces, tight budgets | Bayesian model | High |
| Bisection (Smart Search) | Smooth, independent parameters | Bracket and narrow | High |
| Bisection then TPE Refine | Maximum quality from one run | Narrow, then refine | High |
Want the deep dive on how each one searches? Browse all seven optimization methods.
Choosing on a free-tier budget
With ten cycles a day, sample efficiency matters more than raw reach. That is exactly where a learner earns its place: the TPE (Bayesian) sampler concentrates each cycle on the most promising settings, so a handful of runs go a long way. If your space is small and smooth, Bisection (Smart Search) brackets the peak in a deterministic, resumable way you can stop and continue across days.
When you have room for one strong pass and want maximum quality from it, the Bisection then TPE Refine hybrid narrows broadly, then refines around the peak. Save Brute Force for the genuinely tiny grids where its exhaustive sweep fits inside your budget, and lean on Random or Sequential Improvements as cheap, readable probes rather than your main search.
One more lever costs no cycles at all: the ranges you give the optimizer. Tight, sensible ranges make every method converge faster and waste fewer runs. See how to choose parameter ranges before you spend your daily budget, and study real strategy backtests compared to buy and hold to see disciplined optimization in action.
Brand new to the tool? The getting started walkthrough shows where the method picker lives and runs your first optimization end to end. And if you are still weighing tools, see how to choose a TradingView optimizer against the alternatives before you commit.
Read this first
Optimization is a multiplier, not a magic pill.
Optimization helps a good strategy reach its potential. It cannot invent an edge that was never there. Two things have to be true, in this order, before any optimization is worth running.
Good strategy × right parameters = results
Miss either factor and you are multiplying by zero.
- 1
A strategy with a real edge
Optimization tunes the dials of a strategy. It cannot create signal from noise. If the underlying idea has no edge, no combination of settings will save it: you will only find the parameters that fit the past best and then fall apart live. Garbage in, garbage out. Start with a strategy you understand and believe in.
- 2
The right parameters
Even a genuinely good strategy, run on the wrong inputs, leaves most of its edge on the table, and at worst looks broken when it is not. The default settings are almost never optimal for your market and timeframe. This is the half TradingTune automates: it sweeps the parameter space and surfaces the settings that actually perform.
Bring the edge, we will find the settings. Skip the first half and you are polishing a strategy that was never going to work. Skip the second and you are trading a good idea with the brakes on.
Run any method on your own strategy
Install TradingTune and pick any of the seven methods right inside TradingView's strategy dialog. Free tier, no API keys, data stays local. A free account is required to run.
Add to Chrome, it's free