Mid-West power market lit up as everyone celebrated July 4th 2012
At Amazon we have collected quite a few talented ex-energy traders (recovering). Most of us (me included) we sat on 24 hour electricity trading positions. Electricity cannot yet be stored at scale so it has to be consumed as its produced. Past 20 years regulated wholesale markets have been setup to balance this demand and supply minute by minute. The episode was 2011 when I had just started post my hedge fund career. I bought/sold electricity for clients across the US but mostly in the north-east, mid-west markets which are traditionally higher priced and much volatile. New England winters and constraints in gas supply can spike prices to meteoric levels. In trading, cause-n-effect relationship generally hard to establish but in electricity trading its mostly set on a rule of thumb – look at gas prices at city gate multiply that with a factor [7 to 10] and viola you have the electricity price for the next day, week, maybe even the month. As traders who make their income on margin someone has to be wrong on this math. This is what my fund used to do but all ‘algorithmically’ via ML (more on that later) with human judgment laid on top.
I learned the hard way how ML/AI can’t predict when transformers on the high voltage system will blow up. Early in our journey one of our clients was a large industrial company and they handed over a portion of their portfolio to us to trade on their behalf but they always could over-ride positions we loaded in the market, this was a bright and early (7am) exercise. Remind you all energy price transparency is a recent phenomenon. So get the gas prices we had to call around for that specific client. The gas price quotes that day were high (expected) due to high heat (100 degrees in Indy). So when we ran our algo and forecasted all the hourly prices and created the bids I over-rode the machine and called up my client and told them – we should bid higher than the market – because something is up. They agreed but when market cleared at 2PM (its an auction system) I had a sinking feeling, we didn’t clear – b/c our bids were not high enough despite manually bidding over $100. Remind you this is a $30/MWh average market. The next day when our bids started to settle vs real-time generation it was a mayhem. This was all-time high for real-time energy prices in MISO where instantaneous prices ranged from $1000-$2400. We were losing ‘000s per hour. When the day was done we had lost $100K. It went on for three days straight. Lost over $300K, it wiped out my whole P/L and more importantly it was a failure of the promise of algorithms.
I was young and with limited experience and my algos had failed me and my instincts were not well developed. Electricity markets are not transparent. We did a lot of diagnostics and correction of error (COE) but they were mostly backward looking. This is a very well documented event by FERC. See the FERC docket here (also included below)
Understand the physics of power systems and fuel logistics (gas pipelines). Its tedious and takes years to understand but totally worth it.
Deeply understand market circuit breakers. In electricity markets I learned the hard way – scarcity pricing and when it is called upon.
Risk management. Develop a trading policy that accounts for worst case loss and conditions to trigger it. Algorithms lack judgment, they cant tell you when a transmission line will snap or a power plant will go down so your guardrails have to check.
Clairvoyance equals identifying the signs of black swan events. This paid dividends handsomely during the 2014 Polar Vortex. On a single day we returned our clients multi-million dollars on a relatively small exposure.
Explanation from FERC The events of July 6, 2012 in MISO present a good example of where poor timing can lead to unintended consequences as importers attempt to react to high prices in a neighboring RTO or ISO. Prices were high on that date due to an operating reserve shortage caused by a combination of a 1.7 GW increase in load, the loss of a nearly-600 MW unit, and a 1.5 GW drop in imports from PJM. MISO’s Independent Market Monitor identified the lack of coordinated interchange with PJM as the primary cause of a number of the high-price intervals. In this case, net imports declined because prices were approximately $100/MWh higher in PJM at the time. MISO remained in a shortage for another 30 minutes until imports from PJM responded to the new higher MISO interface price (at this point approximately 20 times higher than PJM). Because imports from PJM lagged the event by 30 minutes, the event was actually ending by the time the imports arrived, and the full effect of the imports was not felt until 2 pm, resulting in an overreaction to the shortage. By 2:10 pm interface prices shifted in response to overreaction, net imports had declined, and PJM prices were higher than in MISO while load was still increasing in MISO, leading to a second shortage.