Interview noteThis page restates a public interview in new words. Dollar results, win rates, and payouts are claims made by the speaker or the host. They are not this desk's record, and they are not a result to expect.
Lewis Trumpeter is 20, and the host introduces him as someone training inside a hedge fund. The caption also renders the name Louiswis. He still describes himself as a retail trader. The route was deliberate. He had been trading the way people learn from videos, felt small next to the managers, and went to the fund to get the information. On a scale of one to ten he scores that experience a nine. It reversed the strategy, the risk, the backtest, and the emotions. The thing that mattered most was the shape of a strategy. The ones he saw were mostly simple. All of them were driven by data and could be tested. Nobody was leaning on a feeling or on the news. Ask about a strategy and they could say what it was expected to do and what it had already done. He took that home: backtests, simulations, and more than one way to decide whether a thing is profitable. The hour is two of those methods, seasonality and the events that behave like a season, and then the test that says the result is not a story. Nothing in it is advice.
Stack the years, then throw out the average if one year did the work
Seasonality, as he teaches it, is the same fact as a toy store in December. Something recurs for a reason, year after year. Charts have the famous versions, a Christmas rally, and the narrower ones, tax dates and the currency conversions a company is forced to make. The whiteboard example is not a model he trades. It is the method. Ten years, each drawn from January to December. The peaks and the lows wander. One stretch of the calendar does not. He puts dates on it. From the 1st of December to the 18th, and in the same breath he also says through the 12th, eight of the ten years rose. That count is a positive expectancy before any average exists. It is a claim that the window is systematic, and he agrees it still has to be tested against randomness.
The seasonal graph is what you get by laying the ten years on top of each other and averaging. The problem is an outlier. Nine years that are barely positive or barely negative, averaged toward zero, plus one year at plus 100 percent, and the graph still points up. The first check is the count. Eight years ended the window positive, so it was not one year. The second check is the size of each year. He invents a spread for the lesson, 0.78 percent, 0.1, 0.3, 0.6, and calls the shape a normal distribution. The average of the positive years, which he then states as 0.76, is the expectancy he would carry into an eleventh year. Not the exact print. A positive outcome, inside a confidence interval. Those percentages are the whiteboard's random numbers. They are not a result from an asset.
The same pass works on gold, oil, and currencies. A logical story, including the Christmas rally, is only a reason to scroll. On TradingView he would drop vertical lines on the 1st and the 18th and walk ten years. A short version of the same exercise, also an illustration, is the 13th of March to the 4th of April: how many of the ten years were negative, and what the expectancy was. Four-day seasons exist. So do many assets. The entry, if he adds nothing else, is the open of the window and the exit is the close. No indicator. He wants the gain distributed across the candles inside the window. One candle that does all the work is a different trade, a single day, in during the morning and out in the evening. A row of days that together make the gain is entered the day before the gains start and exited the day after they end. Any one day inside the row can look like a random walk. Over the window they are supposed to converge. He prefers more than ten years. Ten was the drawing.
The mistake is trusting the average and skipping the days. Eight of ten, with two of those years at plus 10 percent, pulls the expectancy up without any consistency. A line that is flat and then jumps on one candle still finishes positive. It is not the pattern he wants. A slow climb spread across the candles is. The median is more robust to the outlier than the average. Most of the software shows the average. If someone is going to do it properly from the start, he says start with the median.
He still uses the model. It is the front door to data, and it stays on a chart, which matters for people who think in charts. Eight out of ten is a number he can hold. A feeling is not. Before the fund, that was the problem. He did not know the chances.
Train on 2000 to 2015, then make the later years the future
A manual backtest, for him, is not worth doing as the proof. You can misread a bar, and you will fit the rules to things that already happened. Two years of hand-checked history tell him nothing about the next year. The split he was taught is in sample and out of sample. Optimize on a training window. His favorite is 2000 through 2015. Thirteen of fifteen years positive, in the example, is where most people start trading. He wants the model run forward, on 2016 through 2026, on data that never entered the training. That can be done by hand on the chart. It can also be done with a model writing the test. He would still have a person learn it manually first: search for a seasonal curve, take 2000 to 2015 rather than the most recent fifteen years, require something like a normal distribution, then walk 2016, 2017, 2018. Three months of live tinkering is not enough to know, and it wastes the time. He scores this, as the next step for a struggling trader, a nine out of ten. The path is closed or it is open. In sample is the training. Out of sample is whether it survives a future it was not shown. A positive result that is not one day carrying the whole thing is an edge he will call data. Entering on the 1st and leaving on the 18th is hard to bend with a feeling.
The same test applies to events and to days of the week, which he treats as shorter seasons. A Federal Reserve meeting can be tagged. If the market usually rises into it and falls after, that is a pattern around the event, and it can be counted by hand. A payday is the one tied to his own research. On the 16th of the month, in America, payments land in portfolios, and many brokers fund the plans that day. The expectation is that the broker, not the person, buys the S&P. His sketch is the 13th, the 14th, and the 15th, then an up day on the 16th, then a giveback. It is one day and one cause. It is mechanical. It is not a fair-value gap. A company converting currency for a tax statement at the new year is the same kind of force on an exchange rate. A theory that the 16th always prints a large up candle is something the chart can confirm or kill. Day-of-week versions are the lines people already say out loud. Fridays long in gold. Tuesdays long in the Nasdaq, if Tuesday is the day with the positive expectancy. He combines them. They are small trades, and a set of them is a portfolio.
He thinks a retail trader should prefer this family to the concepts that cannot be counted, because a claim like "long on the 16th, out on the 20th" can be proven or dropped in minutes or hours. He is not describing a hand backtest. He wants a yes or a no on expectancy.
A book of small edges, sized by the equity
One strategy is, for him, a bad book. A recession, or a single currency, and the account is dependent on one condition. He wants different assets and different clocks, long windows and short ones, each event tested before it is used. The image is a book. The 16th arrives and that page comes out. A seasonal date arrives and that page comes out. The contributions he draws are deliberately small. One strategy at 4.3 percent. A short that exists only as a hedge, at 0.5 percent, which looks useless alone and useful inside the book. A best one at 6 percent. Not the strategies that claim 30 percent a year. Together they are the portfolio. Each one adds a piece. Signals can say which page is live today.
The chart is TradingView, and the feature he uses is the strategy tester, including the deep backtest, so he is not clicking through history. The standard he wants is whether the rules can be written in words and numbers that another person, or a computer, can repeat. If a computer can execute them, the emotions are out of the fill, and the trader knows what the strategy actually is. He does not write the code himself. He will hand the words to someone who does.
The strategy on the screen is the payday, built for the interview so it would be easy to follow. Calendar days 12 through 18, long, on SPY or on QQQ. QQQ is the Nasdaq ETF, which he treats as the same underlying as the Nasdaq futures. SPY is the stand-in for the ES. He also says the same window showed up on the Nasdaq itself. If it lived on only one symbol he would suspect luck. It does not. The two ETFs are highly correlated, and QQQ moves with more volatility. He would not require both at once. Pick the one that fits. The compounded annual growth he first calls about 5 percent a year, and the table then reads 5.15 for the strategy against 8.67 for buy and hold, on a portfolio of $1 million, with 100 percent of the capital assigned to the strategy used on those days. The objection is that 5 percent loses to holding the index. The answer is the clock. The strategy is in the market 23.24 percent of the time, about a quarter. Buy and hold is always in. The drawdown is smaller. He rescales buy and hold to the same amount of time and gets a compounded rate of 2.61. On that comparison the payday window is ahead, and he calls it nearly double. The idle capital can be used somewhere else. That is the edge he is willing to claim: same money, same time invested, and the dated window beats the rescaled index. For the next month he expects 0.47 percent.
The rule is every month, from the 12th to the 18th, with no other filter. Some months are obviously negative and could be removed. He leaves them in. Selecting the pretty months is how a test gets overfit, and he saves that topic. There is no price target and no stop. The entry is the close of the session before, the 11th, because he wants to be in for the 12th. The exit is the close on the 18th. The day of the interview was a Friday, the 12th was the next session, and the cash market was shut the following day, so the practical entry was that Friday. He learned trading with a tight stop and a 1 percent risk, and he was sure that was correct. A time strategy does not die because price dips. The expectancy lives in the dates. A stop that ejects him because the market went down removes him from a bet that was never about the price level. The fund he describes works in time more often than in price. It depends on the fund.
He will not treat the window as immortal. The last 365 days, on the test he pulls up, are a loss. From 2023 the curve is still positive and no longer impressive. From 2000 it climbed for a long time and then flattened. He reads that as a strategy that got known. He is wary of publishing a very profitable rule for that reason. One video does not kill it. Years of people trading it can. A dying edge can point at the next one. If the crowd is front-running the 12th, the new window to test is the 8th through the 12th, trained on 2000 to 2015 and then released onto later data. TradingView is the picture. The real test, in sample, out of sample, and against randomness, is done elsewhere. A beginner can still delete the hardcoded symbols and run the same dates across other markets. The curve is green or it is not, and there is no year of waiting to see it.
One hundred percent of the allocation, and a second hundred if the book is different
Percent of equity is the risk idea that changed his trading the most. A grandmother's $100 in Apple, bought at $100, rises 10 percent when Apple rises 10 percent. The portfolio is the position. He does the same with a strategy. The whole allocation goes in. Over the last 365 days, 100 percent of equity in this payday book would have damaged the account by 3.14 percent. A 50 percent down day is unlikely and still possible, and he says to be ready for it. A person who simply owns the Nasdaq or the S&P carries a similar price risk. His claim is the better return per unit of time and the smaller drawdown.
Diversification is where the leverage comes from. He is not capped at 100 percent. A second strategy that is not correlated can also take 100 percent, and the account is at 200 percent. One market up 1 percent and the other down 1 percent is just the two markets. He compares it to holding €200 in ordinary shares. The maximum he says he has used is six times, on books that were completely uncorrelated. Double leverage, in that frame, is not much. Some brokers offer 500. That is not where he is going. A day trader, he argues, often has more money in the position than this and cannot see it. Historical drawdown decides the share. A calmer strategy can receive more of the equity. A rougher one is cut if the return still justifies keeping it. Fifty percent of the portfolio is a normal allocation. He also says there is no ceiling if the number is reasonable, including a figure of 1,000 percent. That is his boundary statement, not a recommendation he walks through.
A stop or a target can be tested. Download the trades, try distances, and see which one avoids getting kicked out while keeping the return. Code 0.1, 0.2, 0.3, and brute-force the parameter. The danger is keeping the single best cell. That cell can be the random one. Robustness, whether the neighbors of the best parameter also work, is the check, and he leaves the long version for another conversation. The in-sample window on the screen is SPY from 2000 to 2014 and QQQ from 2000 to 2014. The parameter that came out of that training, without the later years, was the 12th to the 18th. Applied forward, the result is still positive, and it is decaying. Optimizing from today backward is the mistake. A short training window, then a simulated future made of data the rules never saw, is how he gets an expectancy that was not trained on the present.
If the rules cannot be written down, they cannot be tested. He has put other people's strategies into that framework and watched three years of work show no positive expectancy. The beginner version is TradingView's tester. The requirement is the same. Words, then code, then a number.
The kiwi window, the heat map, and a short he will not name
The seasonal example he actually selected is a short in NZDUSD, from the 21st of September to the 2nd of October. He slips once and says the 2nd of August. The window he and the host then confirm is the September date. The average on this one is the median. Trained from 2000 to 2020, the in-sample result is positive. Out of sample, on a shorter recent period, it performed better than it did in sample. Nobody could have predicted that. The expectancy he would carry into the trade is that measured number plus a random miss. He does not expect to land on the number. He expects the sign. He would enter on the 21st with 100 percent of the equity he has assigned, or 200 percent if he wants the expectancy scaled by the money, and he would exit on the 2nd of October. No stop. The market gives what it gives.
Forex is the hardest place he knows to find a real edge, because it does not have the upward drift he is willing to call natural in the Nasdaq. NZD was chosen because the same window, or a close one, was checked across the other currencies and this pair had the best data. The edge also appeared in other pairs. That is how he argues it is not one lucky symbol. A heat map colors the expectancy. Green is the long side of the outcome. Red is usable as a short. A star marks the selected rule, the 21st through the 2nd, on the field he reads as the most robust. Choosing the best out-of-sample cell by itself would invent a new strategy, disconnected from the training. The neighbors matter. The 22nd and the 19th are not bad, so the star is not a single random dot. Moving the entry to the 20th should not destroy the result. Entering on the 19th and leaving on the 3rd should look similar. If one changed day deletes the edge, the rule was fitted to a peak, which is a signal of randomness. Ten outcomes from a coin will include a good one. Picking it without looking at the others is how a random result gets promoted. He built his own TradingView seasonality study because he does not trust the graph. The study is a scanner, an older version, in German, still using an average. It flags the next window: enter the 29th, exit the 6th of October, a 70 percent hit rate, an expected return of 0.4 percent. He would not take that long. About 90 percent of its returns sit around day three, and a pattern carried by one day fails the distribution test he set at the whiteboard.
The short the same tool flags is the one he lingers on. A 100 percent hit rate over the last ten years, and an average of 1.82, with the minus sign removed because he already knows the trade is a short. In on the 16th of September, out on the 8th of October, seventeen trading days, ten out of ten years. The chart shows a drawdown inside the short and then the drop, year after year, back to 2016, which is as far as ten years reaches. Day five is the best day and carries 17 percent of the return, not the whole thing. The median is in the study. Skipping the first day or the second still leaves a positive expectancy. It is not the theoretical best. It is the robustness he wants. He can say all of this in public because a miss is already inside the hit rate. With many tested strategies, a year that does not pay is not an emotional event. The data already said it could fail.
The last strategy he will not name. It is a seasonality plus one indicator, short only, on the main German index, which the caption calls the ducks. Only short trades are coded. On the chart he shows, the strategy makes money when that index drops and still makes money when the index rises, and it beats the benchmark without leverage while being invested about 20 percent of the time. An index with an upward drift, beaten by a short that is usually in cash, is the result he is not allowed to unpack. He was involved in building it at the fund. He says the rule is simple enough to find by playing, and he does not give the indicator or the dates. What he does give is the role. It is another page in the book, a short in a different country, so it can sit beside a 200 percent long in the Nasdaq. Different direction, different market, and a high compounded rate he is willing to use. Stack enough of those pages, each one small and tested, and the portfolio is the thing that gets paid. A career, he thinks, can be built from the approach if it is taken further than one video. Most retail traders never look, because the work is slow, the chart pattern is absent, and the instruction is to stay out of the way of the data.