Fincept Performance

Fincept Performance is the performance-analysis engine behind every perf_* tool. perf_stats gives the full performance report of one or many price series side by side: total return, CAGR, maximum drawdown and Calmar ratio; MTD, 3- and 6-month, YTD, 1-, 3-, 5- and 10-year and since-inception returns; Sharpe and Sortino ratios, mean, volatility, skew, kurtosis and the best and worst period at daily, monthly and yearly frequency; average drawdown, average up and down month and win rates. The other tools give monthly return tables, drawdown episodes, single ratios against a benchmark, return transforms, resampling, annualisation, rolling statistics, portfolio weights, clustering, and the checks of a strategy search. It runs on Fincept’s servers; your agent sends prices or data references and gets back a result envelope with tables and charts.

Fincept Performance tools are free. They count toward the rate limit only; a $fincept reference inside a call is charged like a direct call to that tool.

Modules

Tools are grouped in four modules. Each module is a toolset named perf_<module>, so you can list it or search within it. The Fincept Performance reference lists every tool.

Module	Toolset	Tools
Performance statistics	`perf_stats`	`perf_stats`, `perf_monthly_returns`, `perf_drawdowns`, `perf_metrics`
Return transforms	`perf_returns`	`perf_transform`, `perf_resample`, `perf_annualize`, `perf_frequency`, `perf_rolling`
Portfolio weights and clustering	`perf_weights`	`perf_weights`, `perf_limit_weights`, `perf_random_weights`, `perf_clusters`
Strategy selection checks	`perf_validation`	`perf_deflated_sharpe`, `perf_backtest_overfitting`, `perf_bootstrap`

perf_stats is listed by default. perf_catalog returns the live list of modules and tools from inside your agent.

Inputs

Every tool except perf_annualize, perf_limit_weights and perf_random_weights reads price series, given either way, or both:

Input	Shape	Example
`prices`	A table: one column per series (the column name is the series’ name) and a date column (`date`, `datetime`, `time` or `timestamp`; ISO dates or times). Columns by name or row objects.	`{"date": ["2024-01-02", "2024-01-03"], "SPY": [472.6, 468.8], "TLT": [97.2, 96.1]}`
`series`	Named series, each its own table, joined on their dates. The value column defaults to `close`, `price`, `value` or `nav`; `column` picks another.	`[{"name": "AAPL", "data": {"$fincept": {...}}}]`
`columns`	Use only some of the series.	`["SPY", "TLT"]`
`data_kind`	Weights, clustering and the strategy checks also take periodic returns instead of prices: `"returns"`.	`"returns"`

series is how an agent compares instruments it fetches itself: one data reference per instrument, market_get_candles data at path candles as is.

{
  "series": [
    {"name": "AAPL", "data": {"$fincept": {"tool": "market_get_candles", "arguments": {"symbol": "AAPL", "interval": "1d", "limit": 1260}, "path": "candles"}}},
    {"name": "SPY", "data": {"$fincept": {"tool": "market_get_candles", "arguments": {"symbol": "SPY", "interval": "1d", "limit": 1260}, "path": "candles"}}}
  ],
  "rf": 0.04
}

Dates are required. The statistics are calendar-based (month, year and lookback figures), so every table needs a date column. Each series is measured on its own dates.
Times with a UTC offset (2024-01-02T09:30:00-05:00) are read at the exchange’s own clock: bars at most one a day become that local date, so daylight-saving changes do not shift them; intraday bars become UTC times.
Prices must be above zero; a zero or negative price is refused with the series named.
Periods per year. Annualised figures use periods_per_year when given. Otherwise it is inferred from the dates: from their regular frequency when they have one, else from their median spacing (trading days give 252, weekly data 52, monthly 12). The answer says which was used.

Outputs

Returns, drawdowns, volatility and rates are fractions (0.12 is 12%); ratios are plain numbers. Dated results come back as rows with ISO dates and a line chart; reports come back as tables (statistics in rows, series in columns) with metrics for the headline figures.

Tool	Returns
`perf_stats`	Every statistic of every series, lookback returns, and with several series their return correlations and growth rebased to 100. `rf` (an annual rate) or `rf_prices` (a risk-free index) sets the risk-free return; `start` and `end` narrow the report.
`perf_monthly_returns`	Each series’ return by month and year with the year’s return (months outside its span left empty), best and worst month, share of positive months, average up and down month.
`perf_drawdowns`	The drawdown at every date, the deepest episodes with start, end, length and depth, the maximum and average drawdown, the ulcer index, and the drawdown now with when it began.
`perf_metrics`	Total return, CAGR, maximum drawdown, Calmar, Sharpe and Sortino over `rf`, return/risk, ulcer index and ulcer performance index; with `benchmark` (one of the input series), the information ratio and probabilistic momentum. `annualize: false` gives per-period ratios.
`perf_transform`	`returns`, `log_returns`, `excess_returns` and `price_index` (both from returns), `rebase`, `drawdown`, `winsorize` and `rescale`, at every date.
`perf_resample`	Weekly, monthly, quarterly or yearly observations on their actual last dates (the 29th when the 29th was the month’s last trading day).
`perf_annualize`	Returns earned over given days as annual rates, annual rates on a per-period basis, or the years between two dates.
`perf_frequency`	Each series’ frequency, periods per year and span, and with `lookback` whether it has enough observations over that latest window.
`perf_rolling`	Total return, CAGR, maximum drawdown, Calmar, ulcer index, Sharpe, Sortino, volatility, mean return, skew or kurtosis over a moving window, with its last, highest and lowest values.
`perf_weights`	Inverse-volatility, mean-variance or equal-risk-contribution weights (with optional risk budgets), optionally capped at `limit`, and the resulting portfolio’s annualised return and volatility.
`perf_limit_weights`	Weights capped at a maximum, the excess redistributed in proportion.
`perf_random_weights`	Random portfolios within per-asset bounds that sum to a total, reproducible with `seed`.
`perf_clusters`	Groups of assets by the correlation of their returns (k-means or fast threshold clustering), each with its average inner correlation, and the correlation matrix.
`perf_deflated_sharpe`	The deflated Sharpe ratio of a search’s best strategy and the expected best Sharpe ratio of that many skill-less trials. Give every trial’s prices (it picks the best), or the selected strategy with `trial_sharpe_ratios`.
`perf_backtest_overfitting`	The probability of backtest overfitting by combinatorially symmetric cross-validation over every trial’s returns, the winner’s mean out-of-sample rank and the distribution of rank logits.
`perf_bootstrap`	The bootstrap distribution of a statistic per series: the value on the actual returns, the mean, a confidence interval and the share of resamples above zero, with a histogram.

Limits

Limit	Value
Series per call	500
Rows per table	200,000
`perf_rolling` window	10,000 observations
`perf_bootstrap`	10,000 resamples
`perf_backtest_overfitting`	An even number of blocks from 2 to 20 (16 gives 12,870 splits)
Time	30 to 120 seconds; `perf_rolling` and `perf_bootstrap` 180 seconds, `perf_backtest_overfitting` 300 seconds

Inputs the computation cannot use are refused with the reason: a missing date column, repeated dates or series names, a non-positive price, weight bounds or a cap that cannot sum to 1, a benchmark that is not one of the series.

Example

Prompt

Using Fincept, compare AAPL, MSFT and SPY over the last five years: CAGR, max drawdown, Sharpe at a 4% risk-free rate and each one's worst month, then build equal-risk-contribution weights for the three.

The agent fetches each instrument with market_get_candles, passes them as series to perf_stats and perf_monthly_returns, then calls perf_weights with method: "equal_risk_contribution". More prompts:

Prompt	Tools the agent uses
Show the five deepest drawdowns of the Nifty 50 since 2015 and how long each took to recover.	`market_get_candles`, `perf_drawdowns`
Plot NVDA’s rolling one-year Sharpe ratio.	`market_get_candles`, `perf_rolling`
What is AAPL’s information ratio and probability of outperforming SPY?	`market_get_candles`, `perf_metrics`
I tested 40 parameter sets; here are their returns. Is the best one real?	`perf_deflated_sharpe`, `perf_backtest_overfitting`
How confident can I be in this fund’s Sharpe ratio? Bootstrap it.	`market_get_candles`, `perf_bootstrap`
Group these 30 stocks by how their returns move together.	`market_get_candles`, `perf_clusters`

Get started

Connect a client

Use the server

Account

Modules

Inputs

Outputs

Limits

Example

​Modules

​Inputs

​Outputs

​Limits

​Example

Modules

Inputs

Outputs

Limits

Example