optionstocksmachines.com - Profile | ThreadSky | a Reddit-style client for Bluesky

𝗛𝗙𝗪 𝟱: 𝗦𝗶𝗺𝗽𝗹𝗲 𝗺𝗼𝘃𝗶𝗻𝗴 𝗮𝘃𝗲𝗿𝗮𝗴𝗲𝘀 to forecast S&P 500 sector revenues. Not to be confused with moving average models that use lagged error terms! Individual company error rates are often meaningful different from other models. Ensembling? Stay tuned! optionstocksmachines.substack.com/p/hfw-5-simp...

submitted 4 hours ago • 0 comments

𝗕𝗢𝗧𝗪 𝟰: 𝗛𝗼𝗺𝗲 𝗰𝗼𝘂𝗻𝘁𝗿𝘆 𝗯𝗶𝗮𝘀 𝟯 A long VTI - short emerging market portfolio generated 50 pts of outperformance relative to long-only VTI. Global exposure might be great in theory, but very hard to endure in practice! optionstocksmachines.substack.com/p/botw-4-hom... #python #datascience

submitted 3 days ago • 0 comments

𝗛𝗙𝗪𝗪𝗧 𝟱: 𝗥𝗲𝘃𝗲𝗻𝘂𝗲 𝗯𝘂𝗶𝗹𝗱: Concluding code walk-through to build the revenue dataset. Alpha Vantage issues led us to use Simfin. Clean data and an easy to use API. Highlights effort building robust, minimal error investing dataset optionstocksmachines.substack.com/p/hfwwt-5-re... #python #datascience

submitted 5 days ago • 0 comments

𝗛𝗙𝗪 𝟱: 𝗟𝗲𝘀𝘀 𝗹𝗮𝗴𝘀 𝗮𝗿𝗲 𝗺𝗼𝗿𝗲: A four quarter lag model performs worse than both one quarter lag and naïve models. The naïve model is outperforming more sophisticated algorithms! Moving average models next. Stay tuned! optionstocksmachines.substack.com/p/hfw-5-less... #python #timeseries #datascience

submitted 7 days ago • 0 comments

𝗕𝗢𝗧𝗪 𝟯: 𝗥𝗼𝗹𝗹𝗶𝗻𝗴 𝗮 𝗽𝗮𝗶𝗿: Pairs trading oil services and oil ETFs, OIH and USO, adding a rolling regression to calculate the spread ratio improves out-of-sample performance. Further refinements coming. Stay tuned! optionstocksmachines.substack.com/p/botw-3-rol... #python #datascience #pairstrading

submitted 10 days ago • 0 comments

𝗛𝗙𝗪𝗪𝗧 𝟰: 𝗙𝗶𝗻𝗱𝗶𝗻𝗴 𝗙𝘂𝗻𝗱𝗮𝗺𝗲𝗻𝘁𝗮𝗹𝘀: Tickers assigned, time to build the revenue dataset using the Alpha Vantage API. Quick EDA shows too many problems. Scrubbing your dataset is good for your wealth! optionstocksmachines.substack.com/p/hfwwt-4-fi... #python #datascience

submitted 13 days ago • 0 comments

𝗛𝗙𝗪 𝟯: 𝗡𝗮ï𝘃𝗲 𝗶𝘀, 𝗮𝘀 𝗻𝗮ï𝘃𝗲 𝗱𝗼𝗲𝘀 Minimal differences between linear regression and naïve models. Naïve model has higher error rates for a few individual companies regardless of sector. Key takeaway: which models capture company idiosyncrasies better? optionstocksmachines.substack.com/p/hfw-3-naiv...

submitted 14 days ago • 0 comments

𝗕𝗢𝗧𝗪 𝟮: 𝗛𝗼𝗺𝗲 𝗰𝗼𝘂𝗻𝘁𝗿𝘆 𝗯𝗶𝗮𝘀 𝟮 -- 𝗜𝗻𝗱𝗶𝗰𝗲𝘀 𝗺𝗮𝘁𝘁𝗲𝗿 Following up on prior post, we substitute EFA and EEM for VXUS. Long emerging markets made sense for a while. Long VTI and EEM against 25% short EFA generated the highest cumulative return since 2003. optionstocksmachines.substack.com/p/botw-2-hom...

submitted 17 days ago • 0 comments

submitted 18 days ago • 1 comment

𝗛𝗙𝗪𝗪𝗧 𝟯: 𝗕𝗲𝗶𝗻𝗴 𝗻𝗶𝗰𝗲 𝘁𝗼 𝗼𝘂𝗿 𝗳𝘂𝘁𝘂𝗿𝗲 𝘀𝗲𝗹𝘃𝗲𝘀: Still looking for our lost tickers, we try a third option: Yahoo! A manual search might have been faster. But we want to find a programmatic solution to save time for our future selves! optionstocksmachines.substack.com/p/hfwwt-3-wa... #python #datascience

submitted 20 days ago • 0 comments

𝗛𝗙𝗪 𝟮: 𝗢𝗟𝗦 𝗪𝗼𝗿𝗸𝗵𝗼𝗿𝘀𝗲: Forecasting series starts in earnest using linear regression, the ubiquitous workhorse. Deploying a single period lag with recursive forecasting, results are not too different from the naïve baseline. optionstocksmachines.substack.com/p/hfw-2-ols-... #python #timeseries

submitted 21 days ago • 0 comments

𝗕𝗮𝗰𝗸𝘁𝗲𝘀𝘁 𝗼𝗳 𝘁𝗵𝗲 𝗪𝗲𝗲𝗸 𝟭: New (semi) regular series of quick analyses and backtests. First one on home country bias. Nice theory, not so great in practice. Long VTI/short VXUS ETFs outperforms 11 of the last 14 years! optionstocksmachines.substack.com/p/botw-1-hom... #python #datascience #backtesting

submitted 24 days ago • 0 comments

Cleaning company financial data is hard. First, know accounting to make proper adjustments. Then, assign cleaned data to the correct timestamp to be usable for backtests. Time involved: days for one company. How many new 𝒇𝒊𝒏𝒂𝒏𝒄𝒆 𝑨𝑰 companies have spent the time or have the staff to do this?

submitted 25 days ago • 1 comment

𝗛𝗙𝗪𝗪𝗧 𝟮: 𝗜𝗻 𝘀𝗲𝗮𝗿𝗰𝗵 𝗼𝗳 𝗹𝗼𝘀𝘁 𝘁𝗶𝗰𝗸𝗲𝗿𝘀 We test the Tiingo API to find the tickers missed by Alpha Vantage, continuing our walk-thru of how to build the dataset used in our revenue forecasting series. Each Python code step explained. optionstocksmachines.substack.com/p/hfwwt-2-wa... #python #datascience

submitted 26 days ago • 0 comments

𝗛𝗙𝗪 𝟭: 𝗡𝗮ï𝘃𝗲𝘁é: Time series algorithms for fundamental analysis begins in earnest by establishing a naïve benchmark and showing baseline error rates by S&P 500 sector. Next post will detail how we built the dataset with fully reproducible code! optionstocksmachines.substack.com/p/hfw-1-naiv...

submitted 27 days ago • 0 comments

[𝗛𝗙𝗪𝗪𝗧 𝟭]: 𝗛𝗲𝗹𝗹𝗼 𝗙𝗼𝗿𝗲𝗰𝗮𝘀𝘁𝗶𝗻𝗴 𝗪𝗼𝗿𝗹𝗱 𝗪𝗮𝗹𝗸 𝗧𝗵𝗿𝗼𝘂𝗴𝗵 𝟭 We walk through how we built the dataset we will use for our revenue forecasting series, explaining the Python code for each step using the Alpha Vantage API. optionstocksmachines.substack.com/p/hfwwt-1-he...

submitted 38 days ago • 0 comments

New series on time series forecasting for fundamental analysis. Time to size up how well popular time series packages perform against naïve fundamental analysis. optionstocksmachines.substack.com/p/hello-fore...

submitted 39 days ago • 0 comments

I’m thrilled to share some exciting news! I’ve decided to move Option, Stocks, Machines to Substack. I hope this platform will become an even better hub to attract readers interested in using data science to improve, scale, and automate the investment process! optionstocksmachines.substack.com

submitted 40 days ago • 0 comments

𝗗𝗮𝘆 𝟯𝟬 𝗼𝗳 𝟯𝟬 𝗗𝗮𝘆𝘀 𝗼𝗳 𝗕𝗮𝗰𝗸𝘁𝗲𝘀𝘁𝗶𝗻𝗴! Lessons learned. Targeted tweaks benefit simple strategies. Need a systematic approach for using multiple performance metrics. Place simulations logically in overall analysis. #30daysofbacktesting #python #datascience www.optionstocksmachines.com/post/2024-12...

submitted 51 days ago • 0 comments

𝗗𝗮𝘆 𝟮𝟵 𝗼𝗳 𝟯𝟬 𝗗𝗮𝘆𝘀 𝗼𝗳 𝗕𝗮𝗰𝗸𝘁𝗲𝘀𝘁𝗶𝗻𝗴! Out-of-sample test on four strategies vs. buy-and-hold, the 200-day SMA, and the 60-40 SPY-IEF. Adjusted strategy with error correction most promising if not the highest result #30daysofbacktesting #python #datascience www.optionstocksmachines.com/post/2024-12...

submitted 52 days ago • 0 comments

𝗗𝗮𝘆 𝟮𝟴 𝗼𝗳 𝟯𝟬 𝗗𝗮𝘆𝘀 𝗼𝗳 𝗕𝗮𝗰𝗸𝘁𝗲𝘀𝘁𝗶𝗻𝗴! Additional enhancements boost outperformance by 16% pts vs. the original strategy. But simulations show adjusted strategy still lags behind buy-and-hold. #30daysofbacktesting #python #datascience www.optionstocksmachines.com/post/2024-12...

submitted 53 days ago • 0 comments

Every year I take a few days around New Year's to review articles, blog posts, and pdfs I've saved in my "Read for later" folder. Seems like a lot more this year. Most seem to be going into a "Really read for later" folder. I should probably resolve to do something about that.

submitted 54 days ago • 0 comments

𝗗𝗮𝘆 𝟮𝟳 𝗼𝗳 𝟯𝟬 𝗗𝗮𝘆𝘀 𝗼𝗳 𝗕𝗮𝗰𝗸𝘁𝗲𝘀𝘁𝗶𝗻𝗴! Error correction method refined. Attractive results. Cumulative return outpaces buy-and-hold by 36% pts and the original strategy by 26% pts. The Sharpe ratio jumps to 0.6 www.optionstocksmachines.com/post/2024-12... #30daysofbacktesting #python #datascience

submitted 54 days ago • 0 comments

𝗗𝗮𝘆 𝟮𝟲 𝗼𝗳 𝟯𝟬 𝗗𝗮𝘆𝘀 𝗼𝗳 𝗕𝗮𝗰𝗸𝘁𝗲𝘀𝘁𝗶𝗻𝗴! Five days left! Compare adjusted strategy to original 12-by-12. Statistical tests confirm significant difference in returns for key scenarios. www.optionstocksmachines.com/post/2024-12... #30daysofbacktesting #python #datascience

submitted 55 days ago • 0 comments

𝗗𝗮𝘆 𝟮𝟱 𝗼𝗳 𝟯𝟬 𝗗𝗮𝘆𝘀 𝗼𝗳 𝗕𝗮𝗰𝗸𝘁𝗲𝘀𝘁𝗶𝗻𝗴! Confusion matrices! Adjusted strategy: fewer false positives but more false negatives. Delivers higher average returns on true positives. Misses fewer upticks on false negatives www.optionstocksmachines.com/post/2024-11... #30daysofbacktesting #python #datascience

submitted 56 days ago • 0 comments

𝗗𝗮𝘆 𝟮𝟰 𝗼𝗳 𝟯𝟬 𝗗𝗮𝘆𝘀 𝗼𝗳 𝗕𝗮𝗰𝗸𝘁𝗲𝘀𝘁𝗶𝗻𝗴! More on error correction method. Success due to correcting directional biases in mean-reverting models? Maybe. Lags predicting positive, but beats forecasting negatives returns. www.optionstocksmachines.com/post/2024-11... #30daysofbacktesting #python #datascience

submitted 57 days ago • 0 comments

𝗗𝗮𝘆 𝟮𝟯 𝗼𝗳 𝟯𝟬 𝗗𝗮𝘆𝘀 𝗼𝗳 𝗕𝗮𝗰𝗸𝘁𝗲𝘀𝘁𝗶𝗻𝗴! Logic or luck for enhanced strategy? Bit of both. Adjustment improves performance by capturing forward return direction better. www.optionstocksmachines.com/post/2024-11... #30daysofbacktesting #python #datascience

submitted 58 days ago • 0 comments

𝗗𝗮𝘆 𝟮𝟮 𝗼𝗳 𝟯𝟬 𝗗𝗮𝘆𝘀 𝗼𝗳 𝗕𝗮𝗰𝗸𝘁𝗲𝘀𝘁𝗶𝗻𝗴! 12-by-12 model improves using error correction term, inspired by #ML and the Prognostikon blog. www.optionstocksmachines.com/post/2024-11... Check out Prognostikon: prognostikon.cce.uoa.gr/category/pos... #30daysofbacktesting #python #datascience

submitted 59 days ago • 0 comments

𝗗𝗮𝘆 𝟮𝟭 𝗼𝗳 𝟯𝟬 𝗗𝗮𝘆𝘀 𝗼𝗳 𝗕𝗮𝗰𝗸𝘁𝗲𝘀𝘁𝗶𝗻𝗴! 12-by-12 strategy thus far. Limitations but adding an error correction term to improve trading signals shows promise. www.optionstocksmachines.com/post/2024-11... #30daysofbacktesting #python #datascience

submitted 62 days ago • 0 comments

𝗗𝗮𝘆 𝟮𝟬 𝗼𝗳 𝟯𝟬 𝗗𝗮𝘆𝘀 𝗼𝗳 𝗕𝗮𝗰𝗸𝘁𝗲𝘀𝘁𝗶𝗻𝗴! 3-block circular sampling tests 12-by-12 strategy vs. buy-and-hold. Average underperformance of 11% points. 7-block sampling improves but not enough to make viable as is www.optionstocksmachines.com/post/2024-11... #30daysofbacktesting #python #datascience

submitted 63 days ago • 0 comments

𝗗𝗮𝘆 𝟭𝟵 𝗼𝗳 𝟯𝟬 𝗗𝗮𝘆𝘀 𝗼𝗳 𝗕𝗮𝗰𝗸𝘁𝗲𝘀𝘁𝗶𝗻𝗴! Circular block sampling to gauge outperformance likelihood. Results: 200-day SMA underperforms buy-and-hold by 13% points on average. Outperforms only 25% of the time. www.optionstocksmachines.com/post/2024-11... #30daysofbacktesting #python #datascience

submitted 65 days ago • 0 comments

𝗗𝗮𝘆 𝟭𝟴 𝗼𝗳 𝟯𝟬 𝗗𝗮𝘆𝘀 𝗼𝗳 𝗕𝗮𝗰𝗸𝘁𝗲𝘀𝘁𝗶𝗻𝗴! Exploring simulation methods. Assessing sampling from distributions or historical data. Favor historical block sampling. Use autocorrelation plots to find optimal block size. www.optionstocksmachines.com/post/2024-11... #30daysofbacktesting #python #datascience

submitted 68 days ago • 0 comments

𝗗𝗮𝘆 𝟭𝟳 𝗼𝗳 𝟯𝟬 𝗗𝗮𝘆𝘀 𝗼𝗳 𝗕𝗮𝗰𝗸𝘁𝗲𝘀𝘁𝗶𝗻𝗴! 12-by-12 strategy vs. 200-day SMA. 12-by-12 faces deeper drawdowns after the Tech Bubble, but similar over entire period. Evidence of market regime-driven performance. www.optionstocksmachines.com/post/2024-11... #30daysofbacktesting #python #datascience

submitted 69 days ago • 0 comments

𝗗𝗮𝘆 𝟭𝟲 𝗼𝗳 𝟯𝟬 𝗗𝗮𝘆𝘀 𝗼𝗳 𝗕𝗮𝗰𝗸𝘁𝗲𝘀𝘁𝗶𝗻𝗴! Improved 12-by-12 strategy avoids major market declines and outperforms the 60-40 by 22% points. It also outperforms the 200-day, but that mainly stems from 2015-2018. www.optionstocksmachines.com/post/2024-11... #30daysofbacktesting #python #datascience

submitted 70 days ago • 0 comments

𝗗𝗮𝘆 𝟭𝟱 𝗼𝗳 𝟯𝟬 𝗗𝗮𝘆𝘀 𝗼𝗳 𝗙𝗼𝗿𝗲𝗰𝗮𝘀𝘁𝗶𝗻𝗴! Correcting for snooping. Exploring up-to-date weekly data to enhance trading signals. Long-only outperforms buy-and-hold by 10% points. The Sharpe Ratio is 20% points higher. www.optionstocksmachines.com/post/2024-11... #30daysofbacktesting #python #datascience

submitted 71 days ago • 0 comments

𝗗𝗮𝘆 𝟭𝟰 𝗼𝗳 𝟯𝟬 𝗗𝗮𝘆𝘀 𝗼𝗳 𝗙𝗼𝗿𝗲𝗰𝗮𝘀𝘁𝗶𝗻𝗴! Day 13 looked so great because we intentionally snooped. Correcting for that, performance dropped. We explore balancing data availability and up-to-date models in our next post. www.optionstocksmachines.com/post/2024-11... #30daysofbacktesting #python #datascience

submitted 73 days ago • 0 comments

𝗗𝗮𝘆 𝟭𝟯 𝗼𝗳 𝟯𝟬 𝗗𝗮𝘆𝘀 𝗼𝗳 𝗕𝗮𝗰𝗸𝘁𝗲𝘀𝘁𝗶𝗻𝗴! Unlucky number proves the opposite. First backtest looks great. Too great. Be very skeptical. We reveal why in the next post. Spoiler: introducing a common backtesting error www.optionstocksmachines.com/post/2024-11... #30daysofbacktesting #python #datascience

submitted 74 days ago • 0 comments

𝗗𝗮𝘆 𝟭𝟮 𝗼𝗳 𝟯𝟬 𝗗𝗮𝘆𝘀 𝗼𝗳 𝗙𝗼𝗿𝗲𝗰𝗮𝘀𝘁𝗶𝗻𝗴! 320 training/forecasting combinations for the momentum models reveals that a 5 week training period on 12-week forecasts yields attractive results. www.optionstocksmachines.com/post/2024-11... #30daysofbacktesting #python #datascience

submitted 75 days ago • 0 comments

𝗗𝗮𝘆 𝟭𝟭 𝗼𝗳 𝟯𝟬 𝗗𝗮𝘆𝘀 𝗼𝗳 𝗙𝗼𝗿𝗲𝗰𝗮𝘀𝘁𝗶𝗻𝗴! Significant autocorrelation in 12-by-12 model's residuals at lags 1-7. Iterating train/forecast splits finds combinations with the lowest forecast error trough in the 5-by-1 to 13-by-4 range. www.optionstocksmachines.com/post/2024-11... #30daysofbacktesting

submitted 76 days ago • 0 comments

𝗗𝗮𝘆 𝟭𝟬 𝗼𝗳 𝟯𝟬 𝗗𝗮𝘆𝘀 𝗼𝗳 𝗙𝗼𝗿𝗲𝗰𝗮𝘀𝘁𝗶𝗻𝗴! We examine the 12-by-12 model's residuals. The results are encouraging. The model appears stable within typical return ranges, but residuals may be autocorrelated #30daysofbacktesting #python #datascience www.optionstocksmachines.com/post/2024-11...

submitted 80 days ago • 0 comments

𝗗𝗮𝘆 𝟵 𝗼𝗳 𝟯𝟬 𝗗𝗮𝘆𝘀 𝗼𝗳 𝗕𝗮𝗰𝗸𝘁𝗲𝘀𝘁𝗶𝗻𝗴! Developing a market thesis instead of p-hacking. Applying walk-forward techniques to build successive forecasts. Analyzing residuals. #30daysofbacktesting #python #datascience www.optionstocksmachines.com/post/2024-10...

submitted 81 days ago • 0 comments

𝗗𝗮𝘆 𝟴 𝗼𝗳 𝟯𝟬 𝗗𝗮𝘆𝘀 𝗼𝗳 𝗕𝗮𝗰𝗸𝘁𝗲𝘀𝘁𝗶𝗻𝗴! Baseline effects on the 16 different lookback and look forward momentum models see stability across the same look forward periods. Not surprising in an up trending market. #30daysofbacktesting #python #datascience www.optionstocksmachines.com/post/2024-10...

submitted 82 days ago • 0 comments