The Results of my Deep Research forecast of Tesla’s Share Price and the Canadian Elections
Ray Poynter, 13 May 2025
I think Deep Research is likely to replace quite a few types of research, some of them in the market research area. To investigate more, I tried two simple experiments. The first was to forecast the results of the recent Canadian Federal Elections and the second was to forecast what the closing share price for Tesla would be in two weeks from the forecast. Both forecasts turned out to be interesting, as I will show below.
Forecasting the Tesla Share Price
On 28 April the share price for Tesla was at $285.88. Back in January it had been well over $400, before losing value. In the week leading up to 28 April the share price had gone up every day, from $227.50 to $285.88. Would the price keep rising or would it fall back or perhaps hold at that level?
I asked Deep Research to provide an answer. Specifically, I asked it for the share price on 12 May, a date two weeks away. You can read a long post about how it conducted that research by clicking here. The short version is that it forecast a price of $300 and its logic was that psychologically, in the short-term, $300 was an anchor price. If Tesla’s price fell much below $300, people thought it was too cheap and would buy. If it went much over $300, people would think it was a price bump and would cash in.
As the chart below shows. The forecast was almost perfect but missed in the end. The price for Tesla on 9 May (the last trading day before 12 May) was $298.26. However, that weekend and the Monday saw announcements of trade deals with the UK (specifically including cars) and with China (a major market for Tesla) and its share closed at $318.38.
My conclusion? I think the forecast was interesting. Deep Research indicated prices would be anchored at $300 until there was a market change, and that happened. When there was a market change (the two trade deals, both of which had looked unlikely on 28 April), the forecast was no longer as valid. I am not sure that I would start investing on the basis of suggestions from Deep Research. However, if I had bought shares on 28 April, at $285.88 in the expectation of an increase in price to about $300, it would have worked out OK this time.
The Canadian Election
On 28 April, Canada had a Federal Election. Before voting started, I asked Deep Research for a forecast and then published it in LinkedIn. When I looked at the sources Deep Research searched it was clear that it had mostly referenced the published polls, but it had also looked at the betting markets and the opinions of pundits.
The table below shows the predictions and the latest results from the elections. The predictions for share of votes are pretty good, i.e. they are within the +/-3% error margin that the polling industry usually cites. The main reason for this fit is that the Canadian polling industry did a really good job and that was the main input that Deep Research used.
The forecast for seats won was less good, but in a first-past-the-post country, seats can be won or lost on very small movements and differences. Deep Research said that there was a 65% chance that the Liberals would win a absolute majority, in the end they were two seats short of that, but they were the clear winners.
What are the implications of these two experiments?
Neither experiment, by themselves, indicates very much. Yes, both sets of predictions were interesting and potentially useful. But they could have been just chance. What we need to establish is how likely is a particular case to be useful, and what factors determine its propensity for success and failure.
Key Tip
If you are not using Deep Research, I suggest you start now