Dynamics of link failure events in network markets

摘要

We present a computational model and simulation results on the dynamics of local link failures in markets with network structure. Bandwidth markets are inherently networked, so we focus on telecommunications here. The objective of this paper is to test whether or not network failures will have serious economic consequences. We measure economic consequences by looking at changes in expected bandwidth prices, changes in value-at-risk (VAR) and in conditional-value-at-risk (CVAR). Bandwidth markets may be particularly sensitive to network failures because bandwidth is a non-storable commodity. On the other hand alternative paths with equivalent quality of service (QoS) are perfect substitutes so this may limit sensitivity. Non-storability has contributed to enormous volatility in deregulated electricity prices and observations of enormous price spikes. Bandwidth is a true network commodity in that links in the network itself are the traded commodities. Thus a local failure can affect alternative equivalent paths and this can have a knock-on effect in turn. We used a spot market model incorporating non-storability and alternative path selection on price grounds and limited by QoS-equivalence. Spike models are incorporated based on empirical data. We found that for a realistic large-scale market topology if there are, say, four failures per link per year, half of which are long enough to affect the market, then: expected link prices are increased 12%; VAR is increased by 30%; and CVAR by 40%. This is even with a spike size (x3) that is modest compared to observations in electricity markets (x10-x100). For market participants with capacity positions in such a market these consequences are likely to be serious. Thus if failures occur at this rate their consequence must be included in planning. Furthermore, whilst at low failure intensities the network acts as a dampening factor, at higher intensities it acts as an amplifier and thus cannot be neglected. We believe this amplification to be an emergent phenomenon of any market with network structure, although clearly more important for markets with no storage.