Summary
The personal investment screen. At the top, the model's output for today in two sentences: whether there are stocks on the positive signal list, and whether there is a negative signal on the stocks in your portfolio. The screen does not say buy, sell or hold; it shows model output that is the same for everyone and is not personalised investment advice. Below are your investments, the watch list and the measurements the decision rests on. Investments are stored only in your account; other users cannot see them.
What should I do today?
What it is: The card at the top of the screen: one sentence each for the "Positive signal list" and "Your portfolio".
What data feeds it: It summarises the watch list and the My portfolio table below; it does not compute anything of its own.
What it shows: List side: if there is no watch list, "No positive signal list today" and the reason; if there is one, how many stocks it has and the first trading day the signal is measured from; a warning if the list is stale. Portfolio side: that there are no open investments if so; if there is a negative signal, on which stocks (red); otherwise, that there is none on any of them (green). Investments carrying a caution mark are also named.
What can be concluded from it: "No positive signal list today" is not an error: if no signal in the historical data selected stocks that reliably beat the market in both periods, the application does not make up a list. How the decision was made is explained in the "How was this decided?" section below the watch list.
Watch list
What it is: At most 20 stocks on which the model gave its strongest upward signal on the latest signal day. It is not a buy recommendation but a watch list shown the same to everyone: no end date is given, because no measurable, consistent rise was found in these stocks in the past.
What data feeds it: The candidates are the signal horizons of four models (A · Graph diffusion, B · Cross-sectional ranking, D · Consensus and W · Weighted wave) and three textbook signals computed from prices (M · Momentum: last 12 months, excluding the last month; R · Short-term reversal: last 5 days; L · Low volatility: last 60 days). In model W, a stock lagging behind the companies it is linked to is pulled up, and one running ahead of them is pulled down; the links are built each day only from data up to that day, and all horizons carry the same signal. The list comes from the candidate that meets all of these conditions together: the 20-day difference between the strongest buy and strongest sell buckets is positive in both periods, this difference is too large to be explained by chance in both periods (t ≥ 2 from non-overlapping 20-day windows), and the signal's hit rate on the first tradable day is significant (t ≥ 2). If more than one candidate qualifies, the one with the largest difference averaged over the two periods is chosen; if none qualifies, no list is shown. The list is from this signal's latest LIVE day, otherwise from the latest backtest day. The last price comes from the price_daily table, the consensus state and change activity from the same day's D and C signals, and the wave duration from the propagation equation of A · Graph diffusion.
What it shows: Rank, stock, model direction, confidence, consensus state, change activity, last price, and, according to A · Graph diffusion, how many session hours it takes for a shock in that stock to halve and how many stocks it spreads to. Above the list are the chosen signal, the signal day, the first tradable day and the result of the strongest positive bucket in the two periods. An investment can be entered from each row.
What can be concluded from it: The stocks on the list are those on which the model gives its strongest positive signal. The note above the list states the result of the strongest positive bucket in the two periods; if it was not ahead of the market in both periods, the list is for watching only. If the first tradable day has passed, the pipeline has not run and the list is stale. The decision of which stock to invest in and how much is yours.
Bucket evidence
What it is: What the strongest positive and strongest negative stocks of each model and signal horizon did relative to the market after the first tradable day.
What data feeds it: The backtest signals of the latest analysis run (signal_daily) and daily raw returns (return_daily). Each day the stocks are split into five buckets by signal strength. Because the signal is produced after the close, the return starts from the close of the next trading day, and the difference from the market average of the same days (the equal-weighted return of the stocks that have a signal) is accumulated. The period is split in two, and the two halves are computed separately.
What it shows: Cumulative excess return over 5, 20 and 60 days for bucket 1 (strongest positive) and bucket 5 (strongest negative), in separate columns for the two periods; the "Is the difference reliable?" columns give the t value of the 20-day difference for each period (2 and above in green). The rows that produce the watch list and the negative signal warning are marked. Horizons that produce the same signal day by day are merged into one row.
What can be concluded from it: If a bucket does not have the same sign in both periods, or t is below 2, the result may be chance. For example, a signal that is very strong in one period and close to zero in the other produces no list; the textbook signals are not exempt from this rule. The negative signal comes from a signal whose weakest bucket reliably (t ≤ −2) lagged the market in both periods, and is marked in the table with "Negative signal from this signal". If there is no such signal, a signal whose weakest bucket fell in both periods but not reliably is marked with a "caution mark": a stock falling into this bucket does not count as a negative signal; only an orange caution note appears. Past results do not guarantee the future.
Decay measurement and propagation model
What it is: How a signal's hit rate decays in the days after the signal, and, in the fabric model (A · Graph diffusion), how a shock in one stock spreads to other stocks and how long it takes to die out.
What data feeds it: Decay curve: for each day, how correctly the signal ranks the return k days later across stocks (rank correlation), k = 1 … 60. Propagation model: the propagation matrix W learned by A · Graph diffusion in the latest analysis run; a unit shock given to one stock is carried forward day by day with W. Only links with t of 2 or more in the same direction in both halves of the training period enter W; that is why the shock of most stocks does not pass to any other stock.
What it shows: Decay chart and table: the hit rate on the first day and on the first tradable day, and the status. Wave table: for each stock, the size of its shock remaining in other stocks the next day, the number of stocks it affects, and its half-life (in days and session hours). The overall half-life of the equation is also given.
What can be concluded from it: The waves of A · Graph diffusion halve within a few hours; a signal produced from daily closes cannot catch these waves. Per-stock durations can be computed, but trading on them would need intraday price data. For a signal whose curve is high on the first day and close to zero on the first tradable day, the effect is already over by the time the signal is known.
Volatility: normal range and range breakout
What it is: Each stock's normal range of movement over the next 20 trading days, and what the stocks that moved outside this range did afterwards.
What data feeds it: All daily returns in the return_daily table, split in half down the middle. The next day's volatility is forecast with two equations: EWMA (λ = 0.94) and GARCH(1,1) (by maximum likelihood, with variance targeting of the long-run variance). Both are fitted on the first half and compared on the second half with the QLIKE loss; the one with the lower loss is used. The width of the range is set to cover 95% of the 20-day moves in the first half and tested on the second half. The return after a breakout is measured, as in the bucket evidence, from the first tradable day relative to the market average; the same stock is not counted twice within 20 days.
What it shows: The loss of the two equations and the chosen one, the width of the range (in standard deviations), the actual coverage in the second period, and, for the two kinds of breakout (fell below, rose above), the average return relative to the market over the next 20 days, the t value and the number of events. In the portfolio and the watch list, each stock's range, its daily volatility, and a mark if it moved outside the range in the last 20 days.
What can be concluded from it: The normal range is not a direction forecast; it is for deciding how much money to put in a stock and how large a fall is "normal". A kind of breakout counts as a negative signal only if it significantly (t ≤ −2) lagged the market in both periods. If recoveries followed sharp falls in the past, a fall alone is not counted as a negative signal.
My portfolio
What it is: The investments you entered: open ones with their daily status, sold ones with their realised results.
What data feeds it: Only the records of your account in the holding table. The purchase price is the adjusted close of the purchase date or of the last trading day before it. The current value is computed from the latest adjusted close, and the bucket and negative signals from the latest signal day in the signal_daily table.
What it shows: Amount, purchase price, quantity, current price and value, profit or loss, the stock's current bucket in the model (1 strongest, 5 weakest) and the negative signal status (red: negative signal, green: no negative signal) with its source. Reasons for a negative signal: after the purchase the model placed the stock in the weakest 20% bucket (if this bucket's decline is not reliable, this is not a negative signal but an orange caution mark); change detection raised a downward alarm; the holding date entered in older records has passed; the stock moved outside its normal range in the last 20 days and this kind of breakout was measured to be a negative signal. Each row also shows the stock's 20-day normal range. At the top are the total amount invested, the current value and the profit or loss.
What can be concluded from it: The status is updated as the daily pipeline runs. A negative signal is model output, not a recommendation to sell; if you sold a stock, you close the record with "I sold", so the realised result is kept. Profit and loss are computed with adjusted prices, so they include dividends; they do not include commission or tax.
The investment information, comments and recommendations given here are not within the scope of investment advisory services. Investment advisory services are provided under an investment advisory agreement to be signed between a client and brokerage houses, portfolio management companies, or banks that do not accept deposits. The signals here are produced from historical data with statistical models, are shown the same to everyone and are not personalised; they may not suit your financial situation or your risk and return preferences. Therefore, making investment decisions based solely on the information given here may not produce results that meet your expectations.