{
  "id": "quadratic-variation",
  "title": "Quadratic Variation",
  "lang": "en",
  "source_lang": "en",
  "summary": "Quadratic variation measures the accumulated squared increments of a path and supplies the second-order correction central to stochastic calculus.",
  "category": "mathematics",
  "subcategory": "probability-statistics",
  "type": "concept",
  "aliases": [
    "Square variation"
  ],
  "generated_at": "2026-11-06",
  "model": "pro/gpt-6.1-sol",
  "url": "https://aiwiki.page/en/quadratic-variation/",
  "translations": {
    "zh-Hans": {
      "url": "https://aiwiki.page/zh/quadratic-variation/",
      "json": "https://aiwiki.page/zh/quadratic-variation.json"
    }
  },
  "sections": [
    {
      "heading": "",
      "text": "Quadratic variation is the limiting accumulation of squared increments of a function or stochastic process as a time interval is partitioned increasingly finely. It measures a form of path irregularity that ordinary total variation does not capture. A smooth path has zero quadratic variation, whereas standard Brownian motion has quadratic variation equal to elapsed time. This distinction underlies stochastic calculus, including Itô’s formula, stochastic integration, and the analysis of continuous and jump processes."
    },
    {
      "heading": "Definition",
      "text": "Let \\(X=(X_t)_{t\\geq 0}\\) be a real-valued stochastic process. For a partition\n\n\\[\\pi=\\{0=t_0<t_1<\\cdots<t_m=T\\},\\]\n\nits mesh is\n\n\\[|\\pi|=\\max_{0\\leq i<m}(t_{i+1}-t_i).\\]\n\nThe associated sum of squared increments is\n\n\\[Q_\\pi(X;T) = \\sum_{i=0}^{m-1} \\bigl(X_{t_{i+1}}-X_{t_i}\\bigr)^2.\\]\n\nQuadratic variation is obtained by taking a suitable limit of these sums as the mesh tends to zero. Both the class of processes and the mode of convergence matter: such a limit need not exist for an arbitrary path, and pathwise limits can depend on the chosen sequence of partitions.\n\nThe standard general setting is a right-continuous semimartingale with left limits, defined on a filtered probability space. For deterministic partitions \\(\\pi_n\\) of \\([0,T]\\) whose meshes tend to zero, define\n\n\\[Q_t^{(n)} = \\sum_{i=0}^{m_n-1} \\left( X_{t_{i+1}^{(n)}\\wedge t} - X_{t_i^{(n)}\\wedge t} \\right)^2, \\qquad 0\\leq t\\leq T,\\]\n\nwhere \\(a\\wedge b=\\min(a,b)\\). There is a process \\([X]\\), called the quadratic variation of \\(X\\), such that\n\n\\[\\sup_{0\\leq t\\leq T} \\left|Q_t^{(n)}-[X]_t\\right| \\xrightarrow{\\mathbb P}0.\\]\n\nThis is convergence uniformly on compact time intervals in probability. At each fixed time, it implies convergence in probability of the squared-increment sums.\n\nThe process \\([X]\\) is increasing, right-continuous with left limits, and conventionally starts at \\([X]_0=0\\). Its jump at time \\(t>0\\) satisfies\n\n\\[\\Delta[X]_t=(\\Delta X_t)^2, \\qquad \\Delta X_t=X_t-X_{t-}.\\]\n\nIn particular, if \\(X\\) is continuous, then \\([X]\\) is continuous."
    },
    {
      "heading": "Contrast with total variation",
      "text": "The total variation of a real-valued function \\(f\\) on \\([0,T]\\) is\n\n\\[V_T(f) = \\sup_\\pi \\sum_i |f(t_{i+1})-f(t_i)|.\\]\n\nUnlike quadratic variation, it uses absolute increments and takes a supremum over partitions.\n\nIf \\(f\\) is continuous and has finite total variation, then\n\n\\[\\sum_i |f(t_{i+1})-f(t_i)|^2 \\leq \\left(\\max_i |f(t_{i+1})-f(t_i)|\\right)V_T(f).\\]\n\nBy uniform continuity, the maximum increment tends to zero with the mesh. Hence every continuous finite-variation path has zero quadratic variation.\n\nFor a continuously differentiable function, the same conclusion follows from a simple scaling argument: an increment over a time step of length \\(h\\) is of order \\(h\\), its square is of order \\(h^2\\), and summing approximately \\(T/h\\) such terms gives a quantity of order \\(h\\).\n\nJumps change this conclusion. If \\(A\\) has right-continuous paths with left limits and finite variation on compact intervals, then\n\n\\[[A]_t=\\sum_{0<s\\leq t}(\\Delta A_s)^2.\\]\n\nIts continuous finite-variation component contributes nothing, but each jump contributes its squared size.\n\nThus, zero quadratic variation does not by itself mean that a path is constant. Conversely, nonzero continuous quadratic variation indicates behavior that cannot occur for a continuous finite-variation path."
    },
    {
      "heading": "Brownian motion",
      "text": "For standard Brownian motion \\(B\\),\n\n\\[[B]_t=t\\]\n\nas an identity of processes, up to indistinguishability.\n\nThe basic calculation uses a deterministic partition of \\([0,T]\\). Brownian increments are independent, centered normal random variables with variances equal to their time increments. Consequently,\n\n\\[\\mathbb E\\!\\left[Q_\\pi(B;T)\\right]=T\\]\n\nand\n\n\\[\\operatorname{Var}\\!\\left(Q_\\pi(B;T)\\right) = 2\\sum_i(t_{i+1}-t_i)^2 \\leq 2T|\\pi|.\\]\n\nHere \\(\\mathbb E\\) denotes expectation, and \\(\\operatorname{Var}\\) denotes variance. The variance bound tends to zero, so the sums converge to \\(T\\) in mean square and therefore in probability.\n\nThis is compatible with Brownian paths having infinite total variation on every nontrivial compact time interval, almost surely. Brownian increments over a time interval of length \\(h\\) have a typical magnitude proportional to \\(\\sqrt h\\). Their squares therefore have magnitude proportional to \\(h\\), producing a nonzero limit when accumulated.\n\nFor\n\n\\[X_t=x_0+\\mu t+\\sigma B_t,\\]\n\nwith constant \\(\\mu\\) and \\(\\sigma\\),\n\n\\[[X]_t=\\sigma^2t.\\]\n\nThe constant initial value and continuous finite-variation drift do not affect quadratic variation.\n\nQuadratic variation is not the same as variance. The former is formed from increments along an individual observed path; the latter describes dispersion across a probability distribution. Their numerical equality for standard Brownian motion at time \\(t\\) is a special relationship."
    },
    {
      "heading": "Quadratic covariation",
      "text": "For two semimartingales \\(X\\) and \\(Y\\), quadratic covariation is defined by polarization:\n\n\\[[X,Y]_t = \\frac12\\bigl([X+Y]_t-[X]_t-[Y]_t\\bigr).\\]\n\nEquivalently, it is the limit of sums of products of simultaneous increments:\n\n\\[\\sum_i \\bigl(X_{t_{i+1}\\wedge t}-X_{t_i\\wedge t}\\bigr) \\bigl(Y_{t_{i+1}\\wedge t}-Y_{t_i\\wedge t}\\bigr).\\]\n\nQuadratic covariation is symmetric and bilinear, with\n\n\\[[X,X]=[X].\\]\n\nUnlike quadratic variation, it need not be increasing. Its jumps satisfy\n\n\\[\\Delta[X,Y]_t=\\Delta X_t\\,\\Delta Y_t.\\]\n\nIf \\(A\\) is a continuous finite-variation process, then\n\n\\[[X,A]_t=0.\\]\n\nFor two independent standard Brownian motions \\(B^1\\) and \\(B^2\\),\n\n\\[[B^1,B^2]_t=0.\\]\n\nFor standard Brownian motions with constant correlation \\(\\rho\\),\n\n\\[[B^1,B^2]_t=\\rho t.\\]\n\nFor a vector-valued semimartingale, the collection of covariations forms a matrix\n\n\\[[X]_t=\\bigl([X^i,X^j]_t\\bigr)_{i,j}.\\]\n\nIts increments are positive semidefinite. This matrix records accumulated co-movement rather than the distributional covariance of the process at a single time."
    },
    {
      "heading": "Role in Itô calculus",
      "text": "Product rule\n\nQuadratic covariation supplies the extra term in the stochastic product rule:\n\n\\[X_tY_t = X_0Y_0 + \\int_0^t X_{s-}\\,dY_s + \\int_0^t Y_{s-}\\,dX_s + [X,Y]_t.\\]\n\nThe left limits in the integrands account for possible jumps.\n\nSetting \\(Y=X\\) gives\n\n\\[[X]_t = X_t^2-X_0^2 - 2\\int_0^t X_{s-}\\,dX_s.\\]\n\nThis identity is also a standard characterization of quadratic variation through the Itô integral.\n\nItô’s formula\n\nFor a continuous semimartingale \\(X\\) and a twice continuously differentiable function \\(f\\), Itô’s formula states\n\n\\[f(X_t) = f(X_0) + \\int_0^t f'(X_s)\\,dX_s + \\frac12\\int_0^t f''(X_s)\\,d[X]_s.\\]\n\nThe final integral is the second-order correction absent from the ordinary chain rule for smooth paths.\n\nFor Brownian motion and \\(f(x)=x^2\\), this gives\n\n\\[B_t^2 = 2\\int_0^t B_s\\,dB_s+t.\\]\n\nThe familiar differential notation\n\n\\[(dB_t)^2=dt\\]\n\nis shorthand for the quadratic-variation identity, not an algebraic identity between ordinary infinitesimals.\n\nFor a semimartingale with jumps, the formula becomes\n\n\\[\\begin{aligned} f(X_t) ={}&f(X_0) +\\int_0^t f'(X_{s-})\\,dX_s +\\frac12\\int_0^t f''(X_{s-})\\,d[X]^c_s\\\\ &+\\sum_{0<s\\leq t} \\left[ f(X_s)-f(X_{s-})-f'(X_{s-})\\Delta X_s \\right], \\end{aligned}\\]\n\nwhere \\([X]^c\\) is the continuous part of quadratic variation. Finite jumps require the full correction in the sum, rather than only a second-order approximation.\n\nDiffusion processes\n\nFor an Itô process satisfying the stochastic differential equation\n\n\\[dX_t=b_t\\,dt+\\sigma_t\\,dB_t,\\]\n\nunder the usual integrability assumptions,\n\n\\[[X]_t=\\int_0^t\\sigma_s^2\\,ds.\\]\n\nThus quadratic variation accumulates the squared diffusion coefficient, while the time-integrated drift contributes nothing.\n\nMore generally, if\n\n\\[M_t=\\int_0^t H_s\\,dB_s, \\qquad N_t=\\int_0^t K_s\\,dB_s,\\]\n\nthen\n\n\\[[M,N]_t=\\int_0^t H_sK_s\\,ds.\\]"
    },
    {
      "heading": "Martingales and predictable quadratic variation",
      "text": "For a locally square-integrable local martingale \\(M\\), two related processes must be distinguished:\n\n- Quadratic variation, written \\([M]\\), obtained from squared increments.\n- Predictable quadratic variation, written \\(\\langle M\\rangle\\), the predictable compensator of \\([M]\\).\n\nWith both brackets initialized at zero,\n\n\\[M_t^2-M_0^2-[M]_t\\]\n\nand\n\n\\[M_t^2-M_0^2-\\langle M\\rangle_t\\]\n\nare local martingales. Predictability is defined relative to the underlying filtration.\n\nFor continuous local martingales,\n\n\\[[M]=\\langle M\\rangle.\\]\n\nFor martingales with jumps, they can differ.\n\nFor example, let \\(N_t\\) be a Poisson process of rate \\(\\lambda\\), and set\n\n\\[M_t=N_t-\\lambda t.\\]\n\nEach jump of \\(M\\) has size \\(1\\), so\n\n\\[[M]_t=N_t, \\qquad \\langle M\\rangle_t=\\lambda t.\\]\n\nThe first records the realized number of jumps; the second is its predictable compensator.\n\nIf \\(M\\) is square-integrable and \\(M_0=0\\), then\n\n\\[\\mathbb E[M_t^2] = \\mathbb E[[M]_t] = \\mathbb E[\\langle M\\rangle_t].\\]\n\nThese identities connect quadratic variation to second moments without identifying it with variance path by path."
    },
    {
      "heading": "Pathwise interpretation and applications",
      "text": "Quadratic variation can also be studied for a deterministic function along a specified sequence of partitions. This supports pathwise forms of stochastic calculus, but the partition sequence is part of the construction. The standard probabilistic result for semimartingales does not assert convergence along every possible path-dependent choice of partitions, nor almost-sure convergence along every arbitrary deterministic sequence.\n\nFor a fixed sequence of partitions, convergence almost surely is a stronger statement than convergence in probability and requires separate justification. Brownian motion, for example, admits almost-sure quadratic-variation limits along standard dyadic partitions.\n\nIn financial econometrics, a sum of squared high-frequency returns is called realized variance. Under a semimartingale model and suitable observation conditions, it estimates quadratic variation over the observation interval. For a process with both continuous fluctuations and jumps,\n\n\\[[X]_T=[X]^c_T+\\sum_{0<s\\leq T}(\\Delta X_s)^2.\\]\n\nRealized variance therefore includes squared jumps and need not estimate only integrated diffusion variance. Market microstructure noise and other observation effects can also prevent raw squared-return sums from consistently estimating the latent process’s quadratic variation.\n\nQuadratic variation additionally provides the natural time scale for continuous local martingales. Under the hypotheses of the Dambis–Dubins–Schwarz theorem, such a martingale can be represented as Brownian motion evaluated at the random clock given by its quadratic variation."
    }
  ],
  "links": [
    "stochastic-process",
    "wiener-process",
    "stochastic-calculus",
    "limit",
    "probability-space",
    "convergence-in-probability",
    "total-variation",
    "uniform-continuity",
    "expected-value",
    "variance",
    "almost-surely",
    "probability-distribution",
    "ito-integral",
    "itos-formula",
    "chain-rule",
    "stochastic-differential-equation",
    "martingale",
    "filtration-probability",
    "poisson-process",
    "econometrics"
  ],
  "sources": [],
  "license": "https://creativecommons.org/licenses/by/4.0/"
}