The best and worst trading indicators, ranked across 222,010 backtests. The top 4 are not what you expect.

In November 2025 I ranked every indicator across 114,549 backtests and published what I found. The corpus has since grown to 222,010, so I ran it again. The top of the list is not where most people would look for it.

Previously · I Analyzed 100,000 Backtests to Find the Best Trading Indicator. The Answer Was Not What I Expected

Nothing in the top four is a chart pattern. The best risk-adjusted returns came from strategies that checked their own position before acting. Read that as a fact about the people who write these strategies as much as about the indicators: nobody was randomly assigned a rule, and someone reaching for MaxDrawdown is already thinking about risk. This is a survey of what worked for whoever wrote it, not an experiment.

The reliability floor

What "enough data to rank" means

The corpus contains 93 observed built-in indicator types. Most of them are useless for ranking, because a type used in nine backtests tells you about nine people, not about the indicator.

So there is a reliability floor: a type only enters the ranking once it appears in at least 500 backtests. That cuts 93 down to 34. The other 59 are reported in the full table but never used to make a claim.

Everything is a median Sharpe ratio across traded, hand-authored backtests lasting 365 to 2,000 days. Optimizer-generated runs are excluded, because a genetic search is a population of attempts rather than strategies anyone chose to keep. That exclusion alone removes 74,518 runs.

The ranking

All 34 types that clear the floor

The top four are portfolio state rather than chart signals. They measure how much the portfolio is worth, how far the asset has fallen from its peak, how big the position is, and how far the position is from its cost basis.

Two of those four are closer to one finding than two. PositionValue and PositionPercentChange post an identical 0.927 because they largely appear in the same strategies: 10,407 of the 14,839 backtests using PositionPercentChange also use PositionValue. PortfolioValue and MaxDrawdown are not like that, overlapping in only 72 of 3,150, and both hold their place when I count one observation per portfolio instead of per backtest, at 1.276 and 1.182 across 3,068 and 1,357 distinct portfolios. The top of this table is a top two and a pair.

Show all 93 observed indicator types

Every built-in type observed in the corpus, including the 59 that fall below the 500-backtest floor. Rows in grey are descriptive only: they are too rare to rank fairly, and no claim here rests on them.

All 93 observed built-in indicator types

#	Indicator	Backtests	Median Sharpe
1	PortfolioValue	3,150	1.244
2	SumOrderAmount	4	1.135
3	MaxDrawdown	1,637	1.121
4	PositionValue	31,969	0.927
5	PositionPercentChange	14,839	0.927
6	MaxDrawup	84	0.923
7	CurrentTimeSeconds	2	0.876
8	Multiply	4,547	0.863
9	OptionGrossExposurePercent	1,529	0.859
10	IndicatorExponentialMovingAverage	341	0.852
11	Value	117,723	0.851
12	LastOrderPrice	432	0.769
13	IndicatorStandardDeviation	49	0.755
14	Month	108	0.747
15	DaysSinceLastRebalanceOptionOrder	2,385	0.742
16	OptionDaysHeld	17	0.738
17	PriceRateOfChange	5,132	0.730
18	DaysSinceStrategyFired	934	0.713
19	IndicatorSimpleMovingAverage	1,153	0.673
20	ExponentialMovingAverage	2,137	0.646
21	SimpleMovingAverage	21,634	0.643
22	DaysSinceOrder	44,138	0.640
23	MinutesUntilClose	1,499	0.630
24	Date	206	0.623
25	Price	16,952	0.602
26	OptionPositionCount	2,315	0.600
27	BearishFairValueGap	3	0.581
28	TrailingSum	13	0.537
29	BuyingPower	213	0.532
30	Index	4,168	0.529
31	Day	842	0.455
32	PreviousClosingPrice	9	0.419
33	Divide	724	0.411
34	PriceStandardDeviation	681	0.408
35	Log	11	0.402
36	DaysSinceOptionOrder	1,294	0.402
37	StockReport	111	0.397
38	PriceMeanAbsoluteDeviation	25	0.396
39	AbsoluteValue	67	0.379
40	RelativeStrengthIndex	8,626	0.373
41	Year	37	0.355
42	Plus	850	0.353
43	IndicatorRateOfChange	748	0.347
44	OptionPositionValue	7	0.344
45	Minus	1,598	0.330
46	DaysSinceEarnings	5	0.288
47	DaysSinceAgent	507	0.265
48	PriceChangeSinceOpen	236	0.233
49	MinutesSinceOptionOrder	441	0.224
50	CurrentTimeHours	120	0.205
51	BollingerBand	1,107	0.198
52	OptionPositionPercentChange	218	0.179
53	InitialValue	46	0.175
54	CompoundAnnualGrowthRate	6	0.158
55	ConsecutiveTrue	60	0.145
56	IndicatorMeanAbsoluteDeviation	4	0.137
57	Max	23	0.132
58	IsIndustry	8	0.130
59	Fundamental	618	0.101
60	OptionSpreadCount	4,360	0.094
61	MaximumPrice	243	0.074
62	IsIndexMember	15	0.063
63	Economic	32	0.054
64	MinutesSinceAgent	248	0.054
65	CrossAbove	1,364	0.045
66	CrossBelow	913	0.027
67	OptionDaysToExpiration	27	-0.016
68	CurrentTimeMinutes	152	-0.091
69	PositionMaxDrawdown	962	-0.094
70	OptionUnrealizedPnL	120	-0.113
71	MinimumPrice	184	-0.200
72	GapPercentage	49	-0.210
73	CountTrue	60	-0.336
74	AverageTrueRange	67	-0.449
75	UnderlyingMaxDrawdown	129	-0.536
76	BullishFairValueGap	22	-0.541
77	MinutesAfterOpen	600	-0.571
78	PositionMaxDrawup	50	-0.724
79	VWAP	167	-0.732
80	DaysSinceAlert	53	-0.952
81	DaysUntilEarnings	71	-0.994
82	IndicatorWindowAgo	146	-1.148
83	OptionPositionMaxDrawdown	28	-1.158
84	IsAsset	1	-1.200
85	IndicatorAtMinutesAfterOpen	115	-1.332
86	Negative	24	-1.628
87	DaysSinceTransaction	4	-1.858
88	MinutesSinceOrder	119	-2.724
89	SumOrderQuantity	30	-3.031
90	HighOfDay	9	-4.404
91	OpeningPrice	10	-4.650
92	LowOfDay	4	-5.656
93	Volume	15	-13.744

Where this goes next

What a corpus this size is actually for

A ranking is a survey. Nobody was randomly assigned an indicator, every strategy here was written by a person who chose their own tickers and rules, and people reach for different tools on different problems. The ordering above tells you where to look, not what causes what.

222,010 outcomes is enough to stop guessing and start testing. Three things become possible here that a smaller dataset cannot support.

1. Run the experiment instead of the survey

Take one strategy, one universe, one window, one rebalance cadence, and change exactly one thing. The indicator goes in, or it comes out. Everything else is held. Run both.

That is a controlled experiment, and it is a backtest sweep the platform already knows how to run. The ranking's job is telling you where to point it. You cannot sweep everything, and 34 candidates ordered by median Sharpe is a very good prior.

2. Go down to the configuration, not the type

An indicator type is not one rule, and treating it as one rule is the weakness of every ranking including this one. A 200-day moving average used as a trend filter and a 3-day moving average used as a trigger share a name and nothing else. They belong at opposite ends of this table, and right now they sit in the same row.

I said in the last piece that I could not answer this, because the lookback column was empty across all 304,426 SimpleMovingAverage rows. The column was empty, and the reason was a bug of mine. The corpus exporter looked for a top-level lookback field. The window is stored at window.length, so the extractor found nothing and wrote a null for all 4,424,078 indicator rows. Nothing failed loudly, because the export succeeded and the column existed. It was simply always empty, and the value was sitting in the same table the whole time, one column over in the raw parameters.

Read correctly, it answers the question, and the answer is no.

My first pass at this bucketed every backtest by every window it contained, which counted crossover strategies twice and produced a 0.261 spread I was ready to publish as a finding. Restricted to strategies using a single window, the spread is 0.038 and every pairwise difference straddles zero: a 20-day average beats a 200-day one by 0.045 with an interval of [-0.035, +0.098]. I cannot show you that window length matters.

That is a real answer rather than a failed one, and it cuts both ways. The three neighbouring types are just as unresolved, so the fine structure in the middle of the ranking is mostly noise. What survives is the top of the table, where the intervals are tight and separated, and the bottom, where MinutesAfterOpen is clearly below everything else. Between those, the honest reading is that this survey cannot tell a 50-day average from a 200-day one, and a controlled sweep can.

3. Train on the outcomes

A strategy paired with what it did is a labelled example. 222,010 of them is a training set, and a rare one: the input is structured rather than prose, the label is a number rather than a judgement, and nobody wrote any of it to be training data.

That is the shape you need to fine-tune a model to write strategies that land in the top decile instead of the middle of the distribution. Every language model can already describe what a good strategy looks like. This would be one that has seen 222,010 attempts and what each of them returned. The obvious hazard is that a model trained to maximise backtest Sharpe learns to overfit backtests, so the label has to be out-of-sample performance rather than the number the corpus already holds.

Corrections

What twice the data revised

The corpus doubled, so I checked what it broke. Three conclusions from my 2025 analysis of 114,549 backtests did not survive the larger sample.

A second result

The sell rule

While the ranking was running I split every backtest containing a buy action on whether it also contained a sell action. Strategies that never sold posted a median return of 36.41% against 19.50% for strategies that did. Nearly double, and it holds in both eras rather than being a 2020s bull-market artifact.

Then I restricted the same split to runs carrying a benchmark. Of the strategies that never sold, only 43.3% beat simply holding the asset, with a median alpha of -0.85. Of the strategies that did sell, 57.0% beat it, at +8.77.

Methodology

How this was measured

The source is a snapshot of the NexusTrade backtest archive taken on 6 September 2026, covering 400,203 scanned documents. Every figure above comes from that snapshot under one filter: the backtest was traded, it ran between 365 and 2,000 days, and it was authored by a person rather than the optimizer.

Each backtest counts once per indicator type, so a strategy using the same indicator in four conditions contributes one observation rather than four. Rankings are medians, not means, because a handful of leveraged outliers move a mean and tell you nothing about the typical strategy. Every median carries a 95% bootstrap interval from 4,000 resamples, and where two intervals overlap I do not claim an ordering between them. Window comparisons additionally count only strategies using a single window of that type, because a crossover strategy holds two and would otherwise be counted in both.

Only aggregate results for built-in indicator types appear anywhere in this analysis. Creator-owned custom indicators, their names, their parameters, their identifiers and their underlying strategy logic stay private, and the results-only table they live in cannot be joined back to any strategy, portfolio or user.

Ask the corpus yourself

Every table in this piece is a question I typed in English. The archive is queryable by the platform's AI agent, which writes read-only SQL against the backtest tables and hands back the rows, so "median Sharpe by moving-average window, minimum 500 backtests" is a sentence rather than a query. Ask which setting of an indicator you already use has done best, whether your rule reads better driving a buy or a sell, or how a type performs once you hold the timeframe fixed.

Ask carefully, though, because the agent answers the question you asked. Put that moving-average question to it plainly and it will bucket every backtest by every window it contains, count each 50/200 crossover twice, and handed me a spread I nearly published. Adding "only strategies that use a single window" changes the answer and is the difference between a finding and an artefact. The corpus removes the work of gathering the data. It does not remove the work of asking a clean question.

Every number here comes from the NexusTrade backtest archive, which is why each row carries its sample size instead of a screenshot of a chart. If you want the anonymized aggregate results to run the ranking yourself, send me the word CORPUS.