|
My post from last night noted a system that would trigger long SPY this morning based on the oversold condition we landed on Friday. Well the market has decided to pop a bit early. Rules are rules. The Bear Hunter System was long SPY going in to this weekend and triggered the exit for the todays open. Nice. You can check out the system in the app using the link below. Keep in mind this is the beta. The version going live on the site this week has a more polished interface and some refined indicator logic. https://ift.tt/eXwFN6R Password: bearhunterbeta That beta version will be going away in a couple days. Also, I should note when a signal fires the entry or exit, the execution is for the next days open. For example you can see the “Buy” on 3/6/26 - that enters the next day at the open - $666.39 That system will be available on the systems page soon and for a single lifetime price you get: With a single lifetime price, I believe it is an extremely affordable add-on to most current strategies and portfolios. It is only in the market 17% of the time yet it has an annualized return greater than SPY buy and hold - Think about that.. And in Bear years… I’ll have more detail for that system soon. Happy Trading… Dave Via https://backtest.substack.com/p/market-pop-and-the-system-trigger
0 Comments
My post from last night noted a system that would trigger long SPY this morning based on the oversold condition we landed on Friday. Well the market has decided to pop a bit early. Rules are rules. The Bear Hunter System was long SPY going in to this weekend and triggered the exit for the todays open. Nice. You can check out the system in the app using the link below. Keep in mind this is the beta. The version going live on the site this week has a more polished interface and some refined indicator logic. https://ift.tt/eXwFN6R Password: bearhunterbeta That beta version will be going away in a couple days. Also, I should note when a signal fires the entry or exit, the execution is for the next days open. For example you can see the “Buy” on 3/6/26 - that enters the next day at the open - $666.39 That system will be available on the systems page soon and for a single lifetime price you get: With a single lifetime price, I believe it is an extremely affordable add-on to most current strategies and portfolios. It is only in the market 17% of the time yet it has an annualized return greater than SPY buy and hold - Think about that.. And in Bear years… I’ll have more detail for that system soon. Happy Trading… Dave via Trading Time Machine https://ift.tt/RvVEAeu In Thursday’s post I laid out exactly what I was watching for: RSI(14) below 30, ATRP expanding, the 183 MA still rising. A high-confidence quick strike setup that has gone profitable every single time it has triggered in SPY’s history. Friday closed. The setup triggered. We enter at Monday’s open. Now here’s the interesting wrinkle, and it’s worth understanding before you pull that trigger. Depending on which chart you looked at Friday, the RSI either closed below 30 at 29.86 or held just above it at 30.42. Same ticker. Same 14-period lookback. Two different answers. The culprit: dividend adjustment. Platforms that back-adjust historical prices for dividends subtly rewrite past price action. Every quarterly SPY dividend gets distributed backwards through history, slightly altering the magnitude of prior up and down moves. That changes the average gain/average loss inputs that feed the RSI formula. Usually the difference is meaningless. Sitting right on our threshold of 30, it isn’t. My backtest was built on unadjusted data -- the raw price series, the one that showed 29.86. So by the rules as tested, Friday confirmed the signal. If your charting platform runs on back-adjusted data you saw 30.42 and may have thought, close, but no trigger. This highlights something every rules-based trader needs to nail down before going live: data consistency. Whatever series you tested on, you monitor on. A signal living right on a threshold is exquisitely sensitive to that choice. In this case it’s horseshoes and hand grenades, close enough is close enough, and the system says we’re in. The rest of the checklist is confirmed. RSI has spent over 15 trading days below the 50 level. The 183 MA is rising. ATRP is above 1.5. Monday open. Quick strike. Profit target 1.25%. Maximum hold two days. Rules are rules. Let’s see what the market does. Via https://backtest.substack.com/p/four-times-in-30-years-monday-makes In Thursday’s post I laid out exactly what I was watching for: RSI(14) below 30, ATRP expanding, the 183 MA still rising. A high-confidence quick strike setup that has gone profitable every single time it has triggered in SPY’s history. Friday closed. The setup triggered. We enter at Monday’s open. Now here’s the interesting wrinkle, and it’s worth understanding before you pull that trigger. Depending on which chart you looked at Friday, the RSI either closed below 30 at 29.86 or held just above it at 30.42. Same ticker. Same 14-period lookback. Two different answers. The culprit: dividend adjustment. Platforms that back-adjust historical prices for dividends subtly rewrite past price action. Every quarterly SPY dividend gets distributed backwards through history, slightly altering the magnitude of prior up and down moves. That changes the average gain/average loss inputs that feed the RSI formula. Usually the difference is meaningless. Sitting right on our threshold of 30, it isn’t. My backtest was built on unadjusted data -- the raw price series, the one that showed 29.86. So by the rules as tested, Friday confirmed the signal. If your charting platform runs on back-adjusted data you saw 30.42 and may have thought, close, but no trigger. This highlights something every rules-based trader needs to nail down before going live: data consistency. Whatever series you tested on, you monitor on. A signal living right on a threshold is exquisitely sensitive to that choice. In this case it’s horseshoes and hand grenades, close enough is close enough, and the system says we’re in. The rest of the checklist is confirmed. RSI has spent over 15 trading days below the 50 level. The 183 MA is rising. ATRP is above 1.5. Monday open. Quick strike. Profit target 1.25%. Maximum hold two days. Rules are rules. Let’s see what the market does. via Trading Time Machine https://ift.tt/iLZocY5 Forecast page: https://ift.tt/otdQpXZ The new redesigned forecast gauges are decidedly negative across all the short term periods. Does this mean 100% we are going to get crushed? Certainly the historical matching algo has landed on some days in history where we’ve had some big down days. Are there tradeable bounces inside weak and even bearish markets? Definitely In posts over the last year I’ve highlighted a bullish trade that used the 7 period RSI while above the 200 day moving average. It is a nice setup as markets experience the inevitable pullbacks in bull markets. I like the shorter period RSI for swing trading as it provides a lookback period that typically aligns with those oversold swing bounce areas in bullish markets. As markets move to more bearish or technically ruptured structure the default 14 period RSI begins to shine. In bullish or bearish markets the default 14 period RSI rarely goes full oversold (30) in a bull market or full overbought (70) in a bear market. In bullish markets you’ll notice the 40 level rarely gets penetrated and the same structure in bearish markets emerges as it rarely goes above 60. One of the defining features of a bearish market is the consistent expansion of bar sizes. This increased range across all timeframes creates bars with significantly larger high-to-low percentage spreads. You can visualize this on your own charts using the ATRP indicator. While similar to the standard Average True Range, the ATRP adds a critical layer of context by calculating the specific Percentage of that range relative to price. I track ATRP across all timeframes to develop a proprietary volatility measure, essentially my own version of the VIX. The effects of a bearish regime are visible all the way down to micro-second bars; these bars exhibit expanded ranges just like their daily counterparts. For example, the typical range for a 5-minute bar currently sits at approximately 10 to 11 S&P points. In a quieter, 15 VIX market, those same bars usually fluctuate around 7 points. This expansion across timeframes is a hallmark of the technical shifts I monitor. One of the anomalies of this particular sell-off is that daily ranges have remained relatively compressed. The 5-period daily ATRP currently sits at 1.37%; while elevated, this is notably low compared to a VIX in the mid-20s, which typically demands an ATRP above 1.5%. Historically, when the RSI spends several weeks suppressed below 50, we expect to see much larger bars. Despite the “fear” currently dominating the headlines, the price action suggests a lack of true capitulation. Instead, many traders appear to be quietly positioning for a bounce, betting on the possibility of de-escalation on the war front. By way of comparison, the initial market drawdown in March and April of last year produced ATRP levels significantly higher than what we are seeing today. Even as the market began that “walk down,” the volatility was more pronounced. This contrast highlights the unusual nature of our current environment, where despite the headlines, the actual price action remains relatively contained. The default RSI has now spent 3 weeks (15 trading days) below the 50 level without the RSI level breaching below the oversold 30 level. Again, we are not seeing the fear on the chart, but only in the headlines and a fairly mild walkdown in the market. If and when we do get larger bars and the RSI pegs below the 30 level, there will be an indication of fear in the headlines and the chart. Those types of conditions typically provide opportunity. Usually, it is a short-term play: a high-confidence, quick strike that provides outsized returns relative to time spent in the market. The 183-period moving average. After spending decades diving deep into billions of price bars across every imaginable timeframe, I have found one consistent signal emerging from the data. The 183-period moving average acts as a definitive line of demarcation between two distinct market regimes: the one above it and the one below it. While many studies highlight the 150 or 200-period averages, the 183 offers a much more specific and definable tone across those billions of bars. We will dive deeper into that specific data in a future post. We’ve now had a few days dancing around that moving average. Another element of that moving average is to denote whether it is rising or falling. Currently it is rising. If we get a spike in ATRP and a thrust below 30 with the RSI while still having that rising moving average, backtests have shown quick strikes to the long side can be fruitful. For example if all these conditions are true in SPY:
Buy SPY the next day and place a profit target at 1.25% or exit after 2 days. We’ve only seen this condition trigger on 4 other occasions in the history of SPY since 1993 (we’ll get in to the ones where the 183 MA is declining). With volatility (ATRP) elevated the profit target is hit very quickly. Below you can see those 4 entries and exits on a chart. March 2025 October 2014 June 2012 August 2004 Let’s look at how we handle a declining 183 moving average. While using the same entry criteria, we must apply a tweak to adjust for the change in regime: specifically, a market characterized by a declining moving average and larger bars. We will want to increase the size of our ATRP level to reflect the higher volatility environment. The Rules are the same as above but we increase the ATRP threshold and require a declining moving average:
The exit will reflect the massive volatility and the time typically required to resolve price upward. The profit target is set at 8% and we would hold a maximum of 1 month (21 trading days) These are usually in VERY turbulent times. Headlines will be over the top with fear. But again we exit profitable every time. April 2025 March 2020 March 2009 October 2008 July 2002 September 2001 You should note these are not designed to be high risk-adjusted return type trades. There is, at times, heat to be taken on this type of setup given the highly emotional periods in which they trigger. That being said, there has been price movement ultimately to the upside. You will notice many of these caught longer-term bottoms. Can trades like these fit into your portfolio or trading strategies? That is your call, but these highly emotional setups can be fruitful if you structure them in a way that complements the portfolio structure you have at that time. Have a Great Night! Dave Johnson - Quantitative System Designer at Via https://backtest.substack.com/p/rsi-and-bearish-markets Forecast page: https://ift.tt/otdQpXZ The new redesigned forecast gauges are decidedly negative across all the short term periods. Does this mean 100% we are going to get crushed? Certainly the historical matching algo has landed on some days in history where we’ve had some big down days. Are there tradeable bounces inside weak and even bearish markets? Definitely In posts over the last year I’ve highlighted a bullish trade that used the 7 period RSI while above the 200 day moving average. It is a nice setup as markets experience the inevitable pullbacks in bull markets. I like the shorter period RSI for swing trading as it provides a lookback period that typically aligns with those oversold swing bounce areas in bullish markets. As markets move to more bearish or technically ruptured structure the default 14 period RSI begins to shine. In bullish or bearish markets the default 14 period RSI rarely goes full oversold (30) in a bull market or full overbought (70) in a bear market. In bullish markets you’ll notice the 40 level rarely gets penetrated and the same structure in bearish markets emerges as it rarely goes above 60. One of the defining features of a bearish market is the consistent expansion of bar sizes. This increased range across all timeframes creates bars with significantly larger high-to-low percentage spreads. You can visualize this on your own charts using the ATRP indicator. While similar to the standard Average True Range, the ATRP adds a critical layer of context by calculating the specific Percentage of that range relative to price. I track ATRP across all timeframes to develop a proprietary volatility measure, essentially my own version of the VIX. The effects of a bearish regime are visible all the way down to micro-second bars; these bars exhibit expanded ranges just like their daily counterparts. For example, the typical range for a 5-minute bar currently sits at approximately 10 to 11 S&P points. In a quieter, 15 VIX market, those same bars usually fluctuate around 7 points. This expansion across timeframes is a hallmark of the technical shifts I monitor. One of the anomalies of this particular sell-off is that daily ranges have remained relatively compressed. The 5-period daily ATRP currently sits at 1.37%; while elevated, this is notably low compared to a VIX in the mid-20s, which typically demands an ATRP above 1.5%. Historically, when the RSI spends several weeks suppressed below 50, we expect to see much larger bars. Despite the “fear” currently dominating the headlines, the price action suggests a lack of true capitulation. Instead, many traders appear to be quietly positioning for a bounce, betting on the possibility of de-escalation on the war front. By way of comparison, the initial market drawdown in March and April of last year produced ATRP levels significantly higher than what we are seeing today. Even as the market began that “walk down,” the volatility was more pronounced. This contrast highlights the unusual nature of our current environment, where despite the headlines, the actual price action remains relatively contained. The default RSI has now spent 3 weeks (15 trading days) below the 50 level without the RSI level breaching below the oversold 30 level. Again, we are not seeing the fear on the chart, but only in the headlines and a fairly mild walkdown in the market. If and when we do get larger bars and the RSI pegs below the 30 level, there will be an indication of fear in the headlines and the chart. Those types of conditions typically provide opportunity. Usually, it is a short-term play: a high-confidence, quick strike that provides outsized returns relative to time spent in the market. The 183-period moving average. After spending decades diving deep into billions of price bars across every imaginable timeframe, I have found one consistent signal emerging from the data. The 183-period moving average acts as a definitive line of demarcation between two distinct market regimes: the one above it and the one below it. While many studies highlight the 150 or 200-period averages, the 183 offers a much more specific and definable tone across those billions of bars. We will dive deeper into that specific data in a future post. We’ve now had a few days dancing around that moving average. Another element of that moving average is to denote whether it is rising or falling. Currently it is rising. If we get a spike in ATRP and a thrust below 30 with the RSI while still having that rising moving average, backtests have shown quick strikes to the long side can be fruitful. For example if all these conditions are true in SPY:
Buy SPY the next day and place a profit target at 1.25% or exit after 2 days. We’ve only seen this condition trigger on 4 other occasions in the history of SPY since 1993 (we’ll get in to the ones where the 183 MA is declining). With volatility (ATRP) elevated the profit target is hit very quickly. Below you can see those 4 entries and exits on a chart. March 2025 October 2014 June 2012 August 2004 Let’s look at how we handle a declining 183 moving average. While using the same entry criteria, we must apply a tweak to adjust for the change in regime: specifically, a market characterized by a declining moving average and larger bars. We will want to increase the size of our ATRP level to reflect the higher volatility environment. The Rules are the same as above but we increase the ATRP threshold and require a declining moving average:
The exit will reflect the massive volatility and the time typically required to resolve price upward. The profit target is set at 8% and we would hold a maximum of 1 month (21 trading days) These are usually in VERY turbulent times. Headlines will be over the top with fear. But again we exit profitable every time. April 2025 March 2020 March 2009 October 2008 July 2002 September 2001 You should note these are not designed to be high risk-adjusted return type trades. There is, at times, heat to be taken on this type of setup given the highly emotional periods in which they trigger. That being said, there has been price movement ultimately to the upside. You will notice many of these caught longer-term bottoms. Can trades like these fit into your portfolio or trading strategies? That is your call, but these highly emotional setups can be fruitful if you structure them in a way that complements the portfolio structure you have at that time. Have a Great Night! Dave Johnson - Quantitative System Designer at via Trading Time Machine https://ift.tt/1sMhEn8 I’ve been a total slacker with the forecast page lately, and it’s honestly ridiculous. I’ll do all the heavy lifting: running the matching algo, crunching the data, and syncing the Google Sheets. Then, I’ll just hit a wall. Apparently, the Herculean task of opening the site editor and manually changing the date was just too much for me. It’s the equivalent of architecting a complex system from scratch and then refusing to push the final build to production because I’m tired of looking at the IDE. My bad. While I’m under the hood fixing my own mess, here is a quick reminder: do not mix up the “forecast” with my actual “systems.” They are not the same thing. The forecast is just a historical look at “twin” days to see where we might be headed based on the past. My systems, conversely, are the ones built to survive and thrive in the markets for the long haul, built on actual trade execution and performance. If the gauges happen to align with your own outlook for the timeframe you trade, they are a great tool to have in the toolkit. The gauges also got a redesign and reflect a more modern finance kind of aesthetic. The page can be found here: https://www.tradingtimemachine.com/trading_forecast.html Please bookmark the page and check it each morning. Also speaking of morning, I can’t believe more active traders are not using the Trader Tools page with the voice news and TickStrike to prep for the pre-market with data releases and breaking news throughout the day.. Get over there and check it out. Have a Great Night! Dave Via https://backtest.substack.com/p/i-finally-fixed-the-fking-forecast I’ve been a total slacker with the forecast page lately, and it’s honestly ridiculous. I’ll do all the heavy lifting: running the matching algo, crunching the data, and syncing the Google Sheets. Then, I’ll just hit a wall. Apparently, the Herculean task of opening the site editor and manually changing the date was just too much for me. It’s the equivalent of architecting a complex system from scratch and then refusing to push the final build to production because I’m tired of looking at the IDE. My bad. While I’m under the hood fixing my own mess, here is a quick reminder: do not mix up the “forecast” with my actual “systems.” They are not the same thing. The forecast is just a historical look at “twin” days to see where we might be headed based on the past. My systems, conversely, are the ones built to survive and thrive in the markets for the long haul, built on actual trade execution and performance. If the gauges happen to align with your own outlook for the timeframe you trade, they are a great tool to have in the toolkit. The gauges also got a redesign and reflect a more modern finance kind of aesthetic. The page can be found here: https://www.tradingtimemachine.com/trading_forecast.html Please bookmark the page and check it each morning. Also speaking of morning, I can’t believe more active traders are not using the Trader Tools page with the voice news and TickStrike to prep for the pre-market with data releases and breaking news throughout the day.. Get over there and check it out. Have a Great Night! Dave via Trading Time Machine https://ift.tt/MeWs0A5 I know this is a departure from my usual financial backtesting content, so please bear with me as I share a March Madness tool I’ve been building for a friend. I posted about this tool last night and found some bugs in the outputs. Those should be buttoned up now. Give it a try and let me know what you think. Every March, millions of people fill out brackets based on gut feeling, school loyalty, or whatever mascot they like. This year I did something different. I built a prediction engine powered by real analytics, 24 years of tournament history, and 25,000 simulated tournaments. Here is what it does and how to use it. Test the Model for yourself : https://indexpy-4rakw5m6bpkhnhm7dghoqn.streamlit.app/ What Is It?Dave’s Bracket Madness 2026 is a free web app with the official 2026 bracket fully loaded - all 68 teams. It combines KenPom efficiency rankings, NET rankings, Quad-1 win percentage, strength of schedule, and historical tournament data into a single prediction system. Every calculation runs live in your browser. It is not just a power ranking. It runs 25,000 full tournament simulations and shows you the probability of every team winning the title, reaching the Final Four, and advancing each round - based on how teams with similar profiles have actually performed in March over the last two decades. How the Model WorksThe engine runs in five stages. First it builds a power rating for each team by combining KenPom, NET, Quad-1 record, strength of schedule, and seed. Then it blends in historical seed advancement rates - 12-seeds win 35% of first round games historically, 8-seeds almost never reach the Final Four regardless of how good they are. Then it applies profile multipliers based on traits that predict over or underperformance in March specifically. Then it runs 25,000 full tournament simulations. Then it outputs everything: championship odds, Final Four probabilities, upset targets, a filled bracket, and betting edges. The bracket shown at the bottom of the simulator is the actual result of the last simulation run - not a chalk prediction. Hit Run Again and you get a completely different universe where different upsets happened. That is the point. Six Tools in One AppPower Rankings. All 68 teams ranked by the model’s power rating, with KenPom, NET, SOS, Quad-1 percentage, and - after running the simulation - championship and Final Four probabilities. Sortable by any column. Monte Carlo Simulator. The engine room. Run 25,000 tournaments and see the championship probability for every team. The upset weight slider controls how chaotic each run gets. More on that below. Upset Engine. Shows the eight factors that historically predict upsets - 3pt dependency, turnover pressure, tempo conflict, experience, rebounding, free throw rate, coach track record, and geography. Includes a watchlist of 2026 teams with the strongest upset profiles and a pool expected value table showing why picking underdogs is often the right move. Champion Profile Analyzer. Scores every top-4 seed against the historical champion checklist. About 85% of champions since 2002 rank top-10 in KenPom, top-20 in both offense and defense, and are seeded 1-3. See which 2026 teams match that profile and which ones fall short. Betting Edge Calculator. Select two teams and the model auto-fills their power rating, tempo, and offensive efficiency. Enter the Vegas spread or over/under and it tells you the model’s projection, the implied win percentage, and whether there is a meaningful edge. First round Thursday and Friday is the best window - books are pricing 32 games fast and lines are not sharp yet. My Bracket. Run the simulation, come to this tab, click Generate. A full 64-team bracket fills in automatically based on what just happened in the simulation. Printable. What the Model Knows About MarchThe most interesting part of the model is what 24 years of tournament data actually shows about which teams overperform and underperform their seed. Slow tempo teams are chronically underrated. Virginia, Wisconsin, Saint Mary’s - teams that play in the low 60s possessions per 40 minutes - advance roughly 12% more than their seed predicts. They neutralize better opponents by controlling pace. Virginia’s tempo of 58 this year makes them quietly dangerous. Three-point dependent teams are a coin flip. Alabama takes over 42% of their shots from three. That either means a hot shooting night that beats anyone, or a cold night that loses to anyone. The model treats them as volatile in both directions, which is exactly what the data says. Experience wins in March. Senior-led teams reach the Sweet 16 at a 28% rate. Freshman-heavy teams manage 17%. That gap is enormous and persistent. Michigan State, UConn, Purdue, and Florida all carry high experience ratings. Alabama and Arkansas are on the wrong side of this split. The 8 and 9 seed curse is real. Almost no 8 or 9 seed has ever reached the Final Four regardless of how good they are. The bracket forces them to beat a 1-seed to get there. The model applies a significant penalty to their advancement probability past the Elite Eight. Coaching matters more than most models account for. Izzo, Hurley, Self, Pitino - the data on these coaches in the tournament is persistent and measurable across decades. Michigan State and UConn both carry real coach bonuses in the model. The Upset SliderThis is the most important control in the app. It sits above the Run button in the simulator and goes from Chalk on the left to Chaos on the right. At the middle setting the model runs as designed - some upsets, some variance, championship probabilities that reflect the real balance of power in the field. At 2.5 to 3.0 on the right side, all games compress toward 50/50 and every run produces a genuinely different Final Four. The fan strategy: If you’re a UConn fan, slide it up to 2.5 and keep hitting Run Again until you see them win. That bracket is a legitimately plausible scenario - a world where your team ran hot at the right time. Jump to My Bracket and use it as your pool entry. It’s not cheating the model, it’s exploring a real possible universe. Three Ways to Use ItFor the Bracket Pool
For the Fan Bracket
For Betting
Tune It YourselfThe Model Builder tab lets you adjust how much each metric weighs in the power rating formula. Five presets are included: Default, Analytics Heavy (KenPom at 60%), Committee Thinking (NET and Quad-1 heavy), Defense Wins Titles, and Resume Based. Change the weights, hit Recalculate, and run the simulation again. You get a completely different set of title contenders and a different bracket. What It Does Not KnowThe model uses real metrics for all 68 teams and historical multipliers from actual tournament data. What it does not have: live injury reports, exact KenPom adjusted efficiency numbers (we use rank as a proxy), and real-time tempo data. For a bracket pool it is significantly better than gut picks. For serious betting, plugging in the actual efficiency numbers and current injury status would make it sharper. The simulation variability is real. At the default setting, no single Final Four combination appears in more than about 1-2% of the 25,000 runs. That is March Madness - genuinely unpredictable, even with good data. Good luck. You can find the tool here : https://indexpy-4rakw5m6bpkhnhm7dghoqn.streamlit.app/ Enjoy! Dave via Trading Time Machine https://ift.tt/2hmVHUw I know this is a departure from my usual financial backtesting content, so please bear with me as I share a March Madness tool I’ve been building for a friend. I posted about this tool last night and found some bugs in the outputs. Those should be buttoned up now. Give it a try and let me know what you think. Every March, millions of people fill out brackets based on gut feeling, school loyalty, or whatever mascot they like. This year I did something different. I built a prediction engine powered by real analytics, 24 years of tournament history, and 25,000 simulated tournaments. Here is what it does and how to use it. Test the Model for yourself : https://indexpy-4rakw5m6bpkhnhm7dghoqn.streamlit.app/ What Is It?Dave’s Bracket Madness 2026 is a free web app with the official 2026 bracket fully loaded - all 68 teams. It combines KenPom efficiency rankings, NET rankings, Quad-1 win percentage, strength of schedule, and historical tournament data into a single prediction system. Every calculation runs live in your browser. It is not just a power ranking. It runs 25,000 full tournament simulations and shows you the probability of every team winning the title, reaching the Final Four, and advancing each round - based on how teams with similar profiles have actually performed in March over the last two decades. How the Model WorksThe engine runs in five stages. First it builds a power rating for each team by combining KenPom, NET, Quad-1 record, strength of schedule, and seed. Then it blends in historical seed advancement rates - 12-seeds win 35% of first round games historically, 8-seeds almost never reach the Final Four regardless of how good they are. Then it applies profile multipliers based on traits that predict over or underperformance in March specifically. Then it runs 25,000 full tournament simulations. Then it outputs everything: championship odds, Final Four probabilities, upset targets, a filled bracket, and betting edges. The bracket shown at the bottom of the simulator is the actual result of the last simulation run - not a chalk prediction. Hit Run Again and you get a completely different universe where different upsets happened. That is the point. Six Tools in One AppPower Rankings. All 68 teams ranked by the model’s power rating, with KenPom, NET, SOS, Quad-1 percentage, and - after running the simulation - championship and Final Four probabilities. Sortable by any column. Monte Carlo Simulator. The engine room. Run 25,000 tournaments and see the championship probability for every team. The upset weight slider controls how chaotic each run gets. More on that below. Upset Engine. Shows the eight factors that historically predict upsets - 3pt dependency, turnover pressure, tempo conflict, experience, rebounding, free throw rate, coach track record, and geography. Includes a watchlist of 2026 teams with the strongest upset profiles and a pool expected value table showing why picking underdogs is often the right move. Champion Profile Analyzer. Scores every top-4 seed against the historical champion checklist. About 85% of champions since 2002 rank top-10 in KenPom, top-20 in both offense and defense, and are seeded 1-3. See which 2026 teams match that profile and which ones fall short. Betting Edge Calculator. Select two teams and the model auto-fills their power rating, tempo, and offensive efficiency. Enter the Vegas spread or over/under and it tells you the model’s projection, the implied win percentage, and whether there is a meaningful edge. First round Thursday and Friday is the best window - books are pricing 32 games fast and lines are not sharp yet. My Bracket. Run the simulation, come to this tab, click Generate. A full 64-team bracket fills in automatically based on what just happened in the simulation. Printable. What the Model Knows About MarchThe most interesting part of the model is what 24 years of tournament data actually shows about which teams overperform and underperform their seed. Slow tempo teams are chronically underrated. Virginia, Wisconsin, Saint Mary’s - teams that play in the low 60s possessions per 40 minutes - advance roughly 12% more than their seed predicts. They neutralize better opponents by controlling pace. Virginia’s tempo of 58 this year makes them quietly dangerous. Three-point dependent teams are a coin flip. Alabama takes over 42% of their shots from three. That either means a hot shooting night that beats anyone, or a cold night that loses to anyone. The model treats them as volatile in both directions, which is exactly what the data says. Experience wins in March. Senior-led teams reach the Sweet 16 at a 28% rate. Freshman-heavy teams manage 17%. That gap is enormous and persistent. Michigan State, UConn, Purdue, and Florida all carry high experience ratings. Alabama and Arkansas are on the wrong side of this split. The 8 and 9 seed curse is real. Almost no 8 or 9 seed has ever reached the Final Four regardless of how good they are. The bracket forces them to beat a 1-seed to get there. The model applies a significant penalty to their advancement probability past the Elite Eight. Coaching matters more than most models account for. Izzo, Hurley, Self, Pitino - the data on these coaches in the tournament is persistent and measurable across decades. Michigan State and UConn both carry real coach bonuses in the model. The Upset SliderThis is the most important control in the app. It sits above the Run button in the simulator and goes from Chalk on the left to Chaos on the right. At the middle setting the model runs as designed - some upsets, some variance, championship probabilities that reflect the real balance of power in the field. At 2.5 to 3.0 on the right side, all games compress toward 50/50 and every run produces a genuinely different Final Four. The fan strategy: If you’re a UConn fan, slide it up to 2.5 and keep hitting Run Again until you see them win. That bracket is a legitimately plausible scenario - a world where your team ran hot at the right time. Jump to My Bracket and use it as your pool entry. It’s not cheating the model, it’s exploring a real possible universe. Three Ways to Use ItFor the Bracket Pool
For the Fan Bracket
For Betting
Tune It YourselfThe Model Builder tab lets you adjust how much each metric weighs in the power rating formula. Five presets are included: Default, Analytics Heavy (KenPom at 60%), Committee Thinking (NET and Quad-1 heavy), Defense Wins Titles, and Resume Based. Change the weights, hit Recalculate, and run the simulation again. You get a completely different set of title contenders and a different bracket. What It Does Not KnowThe model uses real metrics for all 68 teams and historical multipliers from actual tournament data. What it does not have: live injury reports, exact KenPom adjusted efficiency numbers (we use rank as a proxy), and real-time tempo data. For a bracket pool it is significantly better than gut picks. For serious betting, plugging in the actual efficiency numbers and current injury status would make it sharper. The simulation variability is real. At the default setting, no single Final Four combination appears in more than about 1-2% of the 25,000 runs. That is March Madness - genuinely unpredictable, even with good data. Good luck. You can find the tool here : https://indexpy-4rakw5m6bpkhnhm7dghoqn.streamlit.app/ Enjoy! Dave Via https://backtest.substack.com/p/march-madness-bracket-simulator-tool |
Dave JohnsonI'm Dave Johnson, a former investment advisor and quantitative system developer with over 30 years of experience building and trading mechanical systems. These days I focus on rules-based research, honest backtests, and sharing what the data actually shows. Archives
August 2026
Categories |
RSS Feed