Partial sum process for records (Q2488470)

Let \(X_{L_1},X_{L_2},\dots,X_{L_n}\) be a sequence of records for i.i.d. \(X_i\) with CDF \(F\). The authors derive a functional CLT for the process \[ Z_n(t)={1\over \psi(n+\sqrt{n})-\psi(n)} \left( \sum_{j=1}^{nt} X_{L_j}-\psi([nt]) \right), \] where \(\psi(x)=F^{-1}(1-\exp(-x))\) in the case where the normalized distribution of \(X_{L_n}\) converges to a normal distribution. It turns out that if \(\psi(x+\sqrt{x})-\psi(x)\) is regularly varying with an index \(\beta\), then \(Z_n\) converges weakly in \(D[0,1]\) to \(t^{\beta-1/2}B(t)\), where \(B(t)\) is a standard Brownian motion.