Monday Nov 30, 2020: Four US common stocks selected by our Chicago Quantum Net Score model (swing trading)
Jeffrey Cohen (email@example.com), Dec 2, 2020
Quick update Dec 4, 2020: Performance Update
Update Dec 10, 2020 (our trading results)
Good morning quantum computing and US equities readers. We have another ‘public’ stock pick (actually 2 pairs of stocks to hold), and a few updates on our quantum algorithm models for you. Not Earth-shattering, but possibly interesting for those who follow quantum computing in financial services.
This past weekend, with the markets closed, we downloaded all current US common stock tickers from the NYSE, NASDAQ Q, and NYSE American (tape b) which make up around 3,400 stocks. We data validated down to 3,153 stocks and ran our models.
We ran the analysis of all 3,153 stocks at once. We calibrate the CQNS score to be zero for the ‘all in’ portfolio, and the zero portfolio. Our job is to find the minimum CQNS score from an equally weighted portfolio, long (no short), unleveraged portfolio of US common stocks.
This time we tried some new things, as follows:
- We ran our analysis on the same data, but on two different hardware systems. We have an iMAC macOS Catalina v10.15.7 (16GB RAM), and an HP Z420 Windows 7 Professional, SP1 server workstation (48GB RAM), which both run Python 3.7.9. We ran the code side by side, and came up with different portfolios.The systems run at different clock speeds, and with different amounts of memory, and therefore run more or less deeply (e.g., in the few hours we set aside to run the experiment on both systems).
- After we run Phase 1 (find the best 64 stock portfolio from 3,153 stocks), we take the union of the two best portfolios found on each system and made that our best pick for phase 2. Phase 2 is where we evaluate those N stocks for the portfolio with the lowest CQNS score, going as deeply as we need to. We typically select a portfolio of between (1,9) stocks. In one system, we had 81 stocks make it to Phase 2 (60 + 21), and in another, 65 stocks passed to Phase 2 (63 + 2). Taking the union of two top portfolios gives us more ‘all stars’ to choose from.
- We ran on the D-Wave Systems Inc. Advantage 1.1 Pegasus (16) system, as opposed to the D-Wave Systems Inc. 2000Q #6 Chimera (16) system, which we typically run on. We did this because both portfolios were over 64 stocks, which is the maximum for Chimera.
- Market conditions have been very positive, and we see higher BETA values in our ‘all-in’ portfolio for the past year. This is driving up expected returns in the market, which makes our CQNS model discount expected returns more aggressively to equal variance. How can that be? Doesn’t BETA always revert to the mean of 1.0? We have 3,153 stocks (many of which are not in the S&P 500), and we compare them to the S&P 500 ETF (SPY) to determine ticker-specific BETA values that we apply against forward-looking market estimates (based on prior year returns, subject to caps and floors). The impact on the model is that we are picking higher BETA stocks this week. As the market falls, even for a few days, the driver for high-BETA stocks falls.
- Our experience on the D-Wave Advantage system was problematic this week. Our first few runs found valid portfolios, but with relatively high CQNS scores. These were better than average, but would not be considered top portfolios. All of a sudden, our runs were finding no valid portfolios. We adjusted parameters, and significantly adjusted our affine transformations of our QUBO, but to no avail. We probably ran about 20 times on the D-Wave unsuccessfully. We did take those quantum solutions into effect in our final genetic algorithm run, so they do influence our solutions, but not sure how helpful they were.
- We picked small capitalization stocks this time with high BETA values. One of the stocks is so thinly traded that it appeared to go an hour without a trade.
- Unlike in our prior runs, the author, Jeffrey Cohen, took a modest trading position in both pairs of stocks over the course of 2 days. These are not significant nor material positions at this time.
- We kept the CQNS formulation the same, but tuned our solvers significantly during this run. Our simulated bifurcator (based on pressure), did an outstanding job selecting stocks in Phase 2. We are learning how to make this an effective solver.
- The two systems reached a large degree of consensus on their respective stock picks across solvers, with the exception of the quantum annealer and the TABU multi-start sampler.
Our stock picks
We selected ($SMLP, $FET) as one portfolio, and ($LPI, $PRTY) as a second portfolio.
Portfolio 1: Summit Midstream Partners, LP (SMLP) and Forum Energy Technologies, Inc. (FET)
Portfolio 2: Laredo Petroleum, Inc. (LPI), and Party City Holdco Inc. (PRTY)
As you can see, these stocks do exhibit some ‘zig and zag’ behavior where one rises where the other falls. We have the counter-balance effect which minimizes losses, but also restricts gains. However, these high BETA stocks are expected to rise and fall assertively with the market.
We spent the past few days improving our core software to run faster (improve the data validation and download process), to tune our solvers, and to format our results better.
We continue to research skewness and kurtosis, as well as quantum walks on graphs, to determine how these metrics and methods can be used to help select attractive stock portfolios.
Update on Dec 4, 2020 (11:45am ET):
These stocks are outperforming their benchmarks, as seen in the chart below. Our unrealized P&L is positive. We plan to hold for ~25 trading days.
One note, SMLP was difficult for us to trade due to lack of liquidity and low volumes. Our buy order (200 shares at market, market open) executed $0.30/share higher than the prior tick.
Another update: This week we invest in our codebase. We add automation, scale the number of US common stocks it analyzes, and better calibrate the solver models. We test the code. We add NASDAQ Global Market(SM) (G) to our stock mix, and pick up more tickers in our FTP pulls. We found Yahoo Finance data download was unavailable twice when market was open.
Dec 10: Our trading results were mixed. We had two stocks up significantly, and three flat (slightly up).
You notice SNDE is an addition to the portfolio. It appears in a subsequent run to go with FET and SMLP. We did see offsetting moves which kept our portfolio value more stable despite individual stock volatility. The real insight was something my brother Seth Cohen told me decades ago: “cut your losses and let your winners soar.” We held LPI as it rose almost 100%.
We also bought more of our slow performers (you see we added to PRTY and FET) to keep our positions balanced. Not sure this was wise as we increased our average share cost. Comments welcome.
As a reminder, we are not registered financial advisors, nor are we offering investment advice. We are pushing ourselves to create quantum algorithms that create quantum advantage, and to improve our skills and expertise in refining models and improving our quantum computing productivity.
Our CQNS formulation, and our data analysis solvers are subject to change without notice, as we continue our research into the stock markets and quantum computing algorithms and solvers.
We continue to improve our art. We now can index our BETA calculations against SPY or QQQ (client choice). We scale up our model to handle another set of NASDAQ stocks (NASDAQ Select — group G) in addition to NASDAQ Global Select — group Q) which brings up our US common stock count to over 4,500.
Finally, we do not have a position in any stock mentioned in this article (we exited for a 16% gain in 9 trading days, net of trading costs).
For more information about our research and methods, please visit our ResearchGate here.
To learn more about our quantum computing consulting start-up, our paid service offerings to select stocks using our platform and algorithm, our offering to apply your ideas on our platform, or our management consulting offerings, please visit our website here.