--- title: "Partial Autocorrelation: Formula" description: "Partial autocorrelation at lag k measures the correlation between a series and its value k periods earlier after removing the linear effect of the..." canonical: https://www.getswoopr.com/indicators/statistical-and-quantitative/partial-autocorrelation/ source: "Swoopr Investment: https://www.getswoopr.com" --- ## Direct Answer > Partial autocorrelation at lag k measures the correlation between a series and its value k periods earlier after removing the linear effect of the intermediate lags. It isolates the direct link at that lag. Values outside roughly plus or minus 1.96 divided by the square root of the sample size are conventionally treated as statistically different from zero. # Partial Autocorrelation: Formula, Meaning, Signals, Examples and How to Use It ## How Is Partial Autocorrelation Calculated? Partial autocorrelation at lag k measures the correlation between a series and its value k periods earlier after removing the linear effect of the intermediate lags. It isolates the direct link at that lag. **Formula:** `Lag 2: phi22 = (r2 - r1^2) / (1 - r1^2), where r1 and r2 are the autocorrelations at lags 1 and 2; higher lags follow the Durbin-Levinson recursion` **Inputs:** - Series values, often returns or changes - Maximum lag - Autocorrelations at each lag **Worked example.** With a lag-1 autocorrelation of 0.6 and a lag-2 autocorrelation of 0.5, the lag-2 partial autocorrelation is 0.219. The inputs are illustrative numbers, not market data. **Reading it.** Values outside roughly plus or minus 1.96 divided by the square root of the sample size are conventionally treated as statistically different from zero. The lag after which the partial autocorrelation cuts off suggests the order of an autoregressive model. **Limits.** Estimates are noisy in short samples, and many lags tested at once will produce some false significant values. It captures only linear dependence. ## What is Partial Autocorrelation? Partial autocorrelation at lag k measures the correlation between a series and its value k periods earlier after removing the linear effect of the intermediate lags. It isolates the direct link at that lag. ## How to use Partial Autocorrelation Values outside roughly plus or minus 1.96 divided by the square root of the sample size are conventionally treated as statistically different from zero. The lag after which the partial autocorrelation cuts off suggests the order of an autoregressive model. A reading alone is never a buy or sell signal; it needs context from price structure, volume, volatility and the broader market regime. ## Strengths - Removing intermediate lags isolates the direct link at lag k, so a spike at one lag is not just carried over from shorter ones. - The cutoff pattern of the partial autocorrelation function is what helps identify the order of an autoregressive model. - Reproducible when the formula and data convention are fixed. ## Weaknesses and limitations - It is typically judged against a band of about two over the square root of N, so in long samples tiny values appear significant. - Each lag's value comes from a regression that gets longer with k, so high-lag estimates use more fitted parameters from the same data. - Correlated indicators can create false confidence by repeating the same underlying information. ## Editorial and risk note This page is educational content, not individualized investment advice. It should not imply guaranteed prediction, accuracy or outperformance. ## Frequently Asked Questions Is Partial Autocorrelation a buy or sell signal? No. Values outside roughly plus or minus 1.96 divided by the square root of the sample size are conventionally treated as statistically different from zero. Estimates are noisy in short samples, and many lags tested at once will produce some false significant values. A trade built on it still needs a trigger, an invalidation rule, a position size and realistic execution assumptions. What is the best setting for Partial Autocorrelation? Partial Autocorrelation is computed from series values, often returns or changes; maximum lag; autocorrelations at each lag, so the choices that matter are the data convention and the window it is read over. Test those across instruments, regimes and out-of-sample periods instead of fitting one historical winner. Can Partial Autocorrelation be used by itself? Estimates are noisy in short samples, and many lags tested at once will produce some false significant values. It captures only linear dependence. For that reason it is read next to measures of direction, participation, valuation, liquidity and risk, not alone. Does Partial Autocorrelation work on every timeframe? The calculation (Lag 2: phi22 = (r2 - r1^2) / (1 - r1^2), where r1 and r2 are the autocorrelations at lags 1 and 2; higher lags follow the Durbin-Levinson recursion) can be run on any bar size, but behavior is not portable: a reading on a five-minute chart describes a very different market window from the same reading on a daily chart. Validate the exact timeframe before relying on it. Why does Partial Autocorrelation give false signals? It is typically judged against a band of about two over the square root of N, so in long samples tiny values appear significant. Each lag's value comes from a regression that gets longer with k, so high-lag estimates use more fitted parameters from the same data. Lag, noise and regime changes add to this, because market participants react to new information after the reading is calculated. How should Partial Autocorrelation be backtested? Rebuild Partial Autocorrelation from series values, often returns or changes using point-in-time data, write the rules before testing, include delisted securities where relevant, model realistic fills and costs, reserve a validation sample and compare with a simpler baseline. S Swoopr Editorial Team The Swoopr Editorial Team produces independent investment education, research, and tools for understanding markets, evaluating opportunities, and managing risk. All content is educational only and does not constitute personalized investment advice. See our [editorial policy](https://www.getswoopr.com/legal/editorial-policy/) and [corrections policy](https://www.getswoopr.com/legal/corrections-policy/).