AI-Models: MPM and 7s. (See TWO charts further down on this page) Markets stable. We remain long, all-in.
[Mar 28, 2024 update: Our AI called it again successfully again. See charts below.] [Mar. 19, 2024 update: I've posted updated results of our statistical AI efforts in charts down the page]

The "Research Log" button has - the research log! You can follow our Perilous Experiments.

With the research log, and our videos, You can follow our Perilous Experiments! Remember: "Vita Brevis, Ars Longa, Ocassio Praeceps, Experimentum Periculosum & Iudicium Difficile." (Trans: Life is short, Mastering the Art takes a Long Time, Opportunity is Fleeting and Falls Away Easily, Experimentation is Dangerous, and Judgement is Difficult.)

Page down to page-bottom, to see our forecasts from our AI, for one of our main holdings. The AI works better than the humans, which is why AI is so popular. Humans are bad investors. They sell at the lows, and buy at the tops. The AI-driven model can do the opposite, since it has no emotions. Simple as that, it is. [Update: New Projection as of March 19st, 2024 - page down to view]

(Note: If you are running an OLD browser, and this button does not work, you can view our Research Log page by clicking on: http://www.gemesysresearch.com/Gemesys_Page_2.html )

More changes. This is our custom-built Python/NumPy based backpropagating neural-network, fabricated using the matrix-math features of Python (works in both 2.7 and 3.x versions), and here shows now a 5000-epoch training run, with completely different dataset. Graphics show messy training that climbs with poro-poro (raining-down) drop-falls. Although the model trains in a messy way, sliding far off the optimum curve, it holds, and trains effectively. We can now get 80% accuracy, using an economically restricted network, and retraining batchs allow overall accuracy to track around 70%, which is good. And we can replicate the results - it is not just due to fortunate initial random weight assignments. (Click image to expand, ESC to exit (or click a hidden X at top right corner of image to close it)).



Would You Like a Little Signal, With Your Randomness?

The results shown here are using completely different datasets. This data is important, and is a tad more harshly random. We continue to develop our custom-made Python/NumPy-only Neural-network programs. See the charts above. We can read and write the trained weights, adjust network learning-rate as training epochs evolve, and check-point and re-start training with different parameters, to enhance fit-accuracy. This allows batched-training, so that we are less likely to train to noise. We have also been using "early-stopping" of network training, and this simple strategy is showing real promise, as evaluation-dataset prediction accuracy shows real improvement, as absolute initial training accuracy is optimized around 60% to 75% level. (This avoids training to "noise".)

Our custom neural-net uses a home-built variant of backpropagation - still training to a sum-squared loss reduction strategy (as opposed to minimizing cross-entropy), and the thing is working well. The image above shows it training to 80% accuracy on our messy dataset. We can push it to over 90%, but then evaluation dataset predictive accuracy falls below the 50% level.

We like this custom-built system approach and are excited by the results. We are still hammering away with TensorFlow, but having something built from scratch is nice, since the dot-product matrix-math is right there, visible, and can be adjusted in various ways. The approach enhances understanding, and suggests real possible opportunity.

We are still running the statistical-model AI (results in charts below). They continue to point upwards. [Mar. 19, 2024 update: I have posted updated results of our statistical AI efforts in the charts below]

Full Disclosure: We STILL remain long and all-in on our bank stocks, telecoms and the mining shares also. Holding onto the Bank shares, has proved to be a less-than-optimal choice. Their performance this year, has been pretty awful. We expected them to recover, and they now seem to be doing so. Curious times, with war, economic uncertainty, bad leaders, and magical technology.

Click on the Research log-entry button for more notes on our "Dismal-Science" view of things.


GEMESYS Ltd. Research & Consulting: We Live in Truth, Offer Enlightenment and if your projects and plans are in trouble, we can help you Change the Program!. Make new choices while you still have freedom of action. Do not wait until your option sets collapse into terminal singularities. Act while you still can.

GEMESYS Ltd Research - *Fiat Lux*

...

[March 28, 2024 - Update] The equity in question that we track with our AI, and that we have a significant position in, crossed the $70 level today, closing at $70.07. The price has tracked as the AI forecast suggested it might do.] [March 19 2024 - Update] Both projection charts here have turned up. The recent reported numbers were good, and the market has responded. The price-data suggests a price-track back to the $70 to $75 level.

Below, is the chart for the higher-frequency 7s model which also now points upwards. We said earlier that we expected a turn from the lower levels, and this has now happened. The AI's are actually quite remarkable, and do a disturbingly good job of forecasting, as they are driven entirely by the market action, and nothing else. They do not get confused or worried, or fearful. They are not swayed by emotional distortion.

This is not to be considered an exercise in investment advice. Make your own decisions, please.

(Full Disclosure: We remain long, all-in, on Canadian bank equities. We view that they represent significant value, especially considering the very stretched valuations in US equities.)


(Click on any Image to expand it, ESC to exit image-expansion.)

[ March 19 2024 update: ] The statistical models have caught and projected the expected turn. This 7s Model matches the M-model. BNS share price has shifted higher. We expect a return into the mid-70's, given recent data improvement, the attractive dividend level ($1.06/shr/qtr), the safety of Canadian banks, the new executive team under new CEO which appears to be getting traction, and general improvement in North and South American economies. Our AI technical model here and the improving economic environment are in alignment. The interest-rate shock change has been absorbed, and rising rates-of-return on capital should benefit most commercial entities. There is some chance that rates may even be reduced.

This is the 7s Model forecast for our testbed security (BNS), which we hold in a couple of portfolios. This model is a variation with very minor difference from the MPM primary model. It dials up slightly the "frequency" of the process the program tries to capture from the data.

The forecast (green line) is now reflecting the possibility of a sustained upturn in share-price.

Full Disclosure: We still remain long and effectively "All-In" on our bank stocks, as dividends are being paid, earnings are reasonable, and operational risks appear to be *MORE* than adequately managed and reserved for.

This information is not "Security or Trading Advice". It is simply an honest reporting of our research efforts.
(Click on image to expand it, and ESC to exit expanded image.)

The difficulty of seeing the forest, for the trees, always remains... The field beyond the forest, yields. But the forest is an attractive nest of chaotic action, held in a fine ecological-orbital balance by the forces of destruction, and re-birth. If you look closely, a small maple tree, with yellow-orange leaves, which are being shed for winter, is visible in the lower-right of this image. In time, if left to nature, these trees with leaves, which are shed in Autumn, and re-created in the Spring, will replace and displace this grove of spruce-trees. But if I keep the deciduous trees cut back, and work to protect the spruce, I can keep this grove intact, and it will grow, and I can enjoy the green spruce branches in deep, cold winter.

Nature is *not* to be left to operate randomly. Our job is to manage and manipulate it, and encourage the natural outcome we wish to have take place. Left to it's own, nature simply kills everything. Humans need to understand this basic biological truth. Our job is to make nature serve us. We must not become confused, and invert this key requirement.