To extend the transmission range governed by limited transmit power, collaborative beamforming (CB) can be employed with a virtual antenna array (VAA) constructed by an unmanned aerial vehicle (UAV) swarm. CB techniques combined with physical layer security (PLS) algorithms are known to better secure transmissions of sensitive information in the presence of eavesdropping attacks, compared to co-located antenna-based schemes. In particular, analog collaborative beamforming (ACB) with random VAA topologies, which is suitable for UAV networks due to its lower signaling/channel estimation overhead compared to the digital CB, can provide a significantly higher secrecy rate by creating noise-like signals at eavesdroppers. However, the increase in energy consumption to improve secrecy rate had been ignored in the existing ACB-based PLS schemes, which is of paramount importance considering the limited onboard energy of UAVs. Thus, motivated by such limitation, in this paper, we examine the security energy efficiency of the ACB-based PLS technique, where a group of UAVs (or VAA elements) randomly change their locations in a distributed manner to randomize the array factor of the VAA, i.e., a stochastic VAA. The proposed stochastic VAA outperforms the existing static VAA in terms of secrecy energy efficiency as verified by analytic and numerical results. Further, the impact of phase errors of the proposed scheme is also presented, which indicates the significance of the accurate phase estimation.