Managing construction risk with weather derivatives

Abstract

Among construction industry participants, weather has been perceived to be one of the most critical factors impacting project cash-flows. The overall impact of weather on the contractor’s project objectives is non-trivial due to construction industry incentive structures and contract specifics. This paper presents a framework that leverages stylized facts from the construction industry to motivate the use of weather derivatives in managing the non-trivial weather impacts. The proposed framework is demonstrated using data provided by a large construction contractor. We show that weather derivative portfolios used for hedging purposes by the contractor can address the contractor’s aversion for losses as well as the complicated relationship between weather and construction. Furthermore, weather derivative hedging reduces the contractor’s incentive to partake in risk chasing behavior in the face of weather delays and reduces the likelihood of the contractor exploiting other claims channels within the project contract.

Notes

1 Hendrickson (1989) states that the project owner has control of the selection of the contract type.

2 In practice, it has been found that multiple lags better describe the auto-correlation structure of the detrended and deseasonalized temperature process (Alexandridis & Zapranis, 2013; Benth & Benth, 2007; Cao & Wei, 2004)

3 “Buttonville A.” is used as a shorthand for “Buttonville Airport” interchangeably throughout the text.

4 Note: For the bin tests, the meters index is normalized by its arithmetic average. Furthermore, the minimum and maximum temperatures in the sample, −13.54${}^{°}$ C and 26.1${}^{°}$ C, do not correspond to the actual daily low and high temperatures in the sample period since the temperature data in our sample only correspond to days when the crews were out working.

5 The first author’s personal experience on the job site motivates the second consideration.

6 To estimate the p-values, the Newey West Adjustment was applied with k = 6 lags as determined by the rule of thumb $k=0.75\times {(\text{dataset}\mathrm{ }\text{size})}^{\frac{1}{3}}$ (Stock & Watson, 2011)

7 In this article, to combat selection bias, any model selection criterion is applied from the first half of the dataset and any inferences are made from the second half of the dataset. Lee et al. (2016) describe this approach as the most straightforward way to avoid selection bias

8 Lee et al. (2016), find that if the training data are used to select a model, then the selected model may exhibit overfitting. At this stage, overfitting is ignored as several other researchers have validated the yearly seasonalities present in the transition probabilities (Lapez Cabrera et al., 2013; Li et al., 2007; Stowasser, 2011)

9 Recall, that the meters completed model is based upon the de-trended meters drilled.

10 Of course, the assumed scope of work is the completion of directional drilling, to match the underlying meters drilling model.

11 $\widehat{(\hspace{0.5em}})$ is used to denote a finite-sample approximation of the expected value of the quantity of interest.

12 In practice, construction firms naturally reduce the winter effects by hiring seasonally and planning jobs around the winter months (Russell & Pilot, 1969). Although not relevant to our single-crew hedging analysis, the contractor has control over the number of crews they hire over a specific period. For example, hiring three crews in the summer months and one crew in the winter months would have a different risk profile than hiring two crews year-round. Naturally, in the former case, one would expect a larger allocation in the CDD indices.

13 In most of the literature, VaR is defined with a confidence level β above 0.9. In this work, $\beta =0.7,0.8$ are shown simply for illustration of the sensitivity analysis.

Additional information

Notes on contributors

David Islip

David Islip is a PhD student in Industrial Engineering at the University of Toronto under the supervision of Professor Roy H. Kwon. As a part of his research, David also works at Canada Guaranty Mortgage Insurance Company as a Quantitative Risk Analyst performing strategy and risk analytics. His research interests are data-driven decision-making and risk management. He holds a B.ASc in the Engineering Science Infrastructure Option from the University of Toronto.

Jason Z. Wei

Jason Wei is Professor of Finance at the University of Toronto, which he joined in 1998. Prior to 1998, he was Associate Professor of Finance at the University of Saskatchewan. He earned his B.Sc. from Harbin Institute of Technology, MBA from York University, and PhD from the University of Toronto. His research is on empirical asset pricing. He served as the Finance Division Editor of the Canadian Journal of Administrative Sciences from 2005 to 2011, and he is currently an Associate Editor of The Journal of Derivatives.

Roy H. Kwon

Roy H. Kwon is Associate Professor of Operations Research in the Department of Mechanical … Industrial Engineering at the University of Toronto and a member of the faculty in the Masters of Mathematical Finance program (MMF). He received his PhD from the University of Pennsylvania and currently serves as associate editor for several journals in the areas of financial optimization and capital investment. He is main research areas are in portfolio optimization and asset allocation, risk management, and option pricing and he is also active in consulting in the financial industry.