Smooth distribution function estimation for lifetime distributions using Szasz–Mirakyan operators

Abstract

In this paper, we introduce a new smooth estimator for continuous distribution functions on the positive real half-line using Szasz–Mirakyan operators, similar to Bernstein’s approximation theorem. We show that the proposed estimator outperforms the empirical distribution function in terms of asymptotic (integrated) mean-squared error and generally compares favorably with other competitors in theoretical comparisons. Also, we conduct the simulations to demonstrate the finite sample performance of the proposed estimator.

Appendix

The following theorem can be found in Ouimet (2020). He pointed out a mistake in the paper of Leblanc (2012) which also has an impact on this paper. The asymptotic behavior of $R_{1,m}^S$ in Lemma 3 has been corrected compared to Lemma 3 in Hanebeck and Klar 2020, arXiv v.1.

Theorem 8

We define

$$\begin{aligned} V_{k,m}(x)=\frac{(mx)^k}{k!}e^{-mx}, \; \phi (x)=\frac{1}{\sqrt{2\pi }}e^{-x^2/2}, \; and \; \delta _k=\frac{k-mx}{\sqrt{mx}}. \end{aligned}$$

Pick any $\eta \in (0,1)$. Then, we have uniformly for $k \in {\mathbb {N}}_0$ with $\left| \frac{\delta _k}{\sqrt{mx}}\right| \le \eta$ that

$$\begin{aligned} \frac{V_{k,m}(x)}{\frac{1}{\sqrt{mx}}\phi (\delta _k)}&=1+m^{-1/2}\frac{1}{\sqrt{x}}\left( \frac{1}{6}\delta _k^3-\frac{1}{2}\delta _k\right) \\&+m^{-1}\frac{1}{x}\left( \frac{1}{72}\delta _k^6-\frac{1}{6}\delta _k^4+\frac{3}{8}\delta _k^2-\frac{1}{12}\right) +O_{x,\eta }\left( \frac{|1+\delta _k|^9}{m^{3/2}}\right) \end{aligned}$$

as $n \rightarrow \infty$.

We now present various properties of $V_{k,m}$ that are needed for the proofs. The following lemma and its proof are similar to Lemma 2 and Lemma 3 in Leblanc (2012). As mentioned before, parts (e) and (h) take the suggestions in Ouimet (2020) into account. The proofs for these parts are adjusted accordingly.

Lemma 3

Define

$$\begin{aligned}&L_m^S(x)=\sum _{k=0}^{\infty }V_{k,m}^2(x), \\&R^S_{j,m}(x)=m^{-j}\mathop {\sum \sum }_{0\le k < l \le \infty }(k-mx)^jV_{k,m}(x)V_{l,m}(x) {\text { for }} j\in \{0,1,2\}, \end{aligned}$$

and

$$\begin{aligned} \tilde{R}_{1,m}^S(x)=m^{1/2}\sum _{k,l=0}^{\infty }\left( \frac{k\wedge l}{m}-x\right) V_{k,m}(x)V_{l,m}(x), \end{aligned}$$

and $V_{k,m}(x)=e^{-mx}\frac{(mx)^k}{k!}$. It trivially holds that $0 \le L_m^S(x) \le 1$ for $x\in [0,\infty )$. In addition, the following properties hold.

• (a) $L_m^S(0)=1$ and $\displaystyle \lim _{x\rightarrow \infty } L_m^S(x)=0$,

• (b) $R_{j,m}^S(0)=0$ for $j\in \{0,1,2\}$,

• (c) $0 \le R_{2,m}^S(x) \le \frac{x}{m} {\text { for }} x \in (0,\infty )$,

• (d) $L_m^S(x)=m^{-1/2}\left[ (4\pi x)^{-1/2}+o_x(1)\right] {\text { for }} x\in (0,\infty )$,

• (e) $\tilde{R}_{1,m}^S(x)= -\sqrt{\frac{x}{\pi }}+o_x(1) {\text { for }} x\in (0,\infty )$ and $R_{1,m}^S(x)=m^{-1/2}\left[ -\sqrt{\frac{x}{4\pi }}+o_x(1)\right]$,

• (f) $m^{1/2} \displaystyle \int _0^{\infty } L_m^S(x)e^{-ax}{\mathrm {d}}x =\frac{1}{2\sqrt{a}}+o(1)$ for $a \in (0,\infty )$,

• (g) $m^{1/2} \displaystyle \int _0^{\infty } x L_m^S(x)e^{-ax}{\mathrm {d}}x =\frac{1}{4a^{3/2}}+o(1)$ for $a \in (0,\infty )$,

• (h) For any continuous and bounded function g on $[0,\infty )$, $m^{1/2} \displaystyle \int _0^{\infty } g(x)R_{1,m}^S(x)e^{-ax}{\mathrm {d}}x = -\displaystyle \int _0^{\infty } g(x)\frac{\sqrt{x}}{\sqrt{4\pi }}e^{-ax}{\mathrm {d}}x+o(1)$ for $a \in (0,\infty )$ and $\displaystyle \int _0^{\infty } g(x)\tilde{R}_{1,m}^S(x)e^{-ax}{\mathrm {d}}x = -\displaystyle \int _0^{\infty } g(x)\frac{\sqrt{x}}{\sqrt{\pi }}e^{-ax}{\mathrm {d}}x+o(1)$.